remote sensing

Article
Using LiDAR System as a Data Source for Agricultural
Land Boundaries
Natalia Borowiec * and Urszula Marmol

Department of Photogrammetry Remote Sensing of Environment and Spatial Engineering, Faculty of Mining
Surveying and Environmental Engineering, AGH University of Science and Technology, Al. Mickiewicza 30,
30-059 Krakow, Poland; [email protected]
* Correspondence: [email protected]

Abstract: In this study, LiDAR sensor data were used to identify agricultural land boundaries. This is
a remote sensing method using a pulsating laser directed toward the ground. This study focuses on
accurately determining the edges of parcels using only the point cloud, which is an original approach
because the point cloud is a scattered set, which may complicate finding those points that define the
course of a straight line defining the parcel boundary. The innovation of the approach is the fact that
no data from other sources are supported. At the same time, a unique contribution of the research is
the attempt to automate the complex process of detecting the edges of parcels. The first step was to
classify the data, using intensity, and define land use boundaries. Two approaches were decided, for
two test fields. The first test field was a rectangular shaped parcel of land. In this approach, pixels
describing each edge of the plot separately were automatically grouped into four parts. The edge
description was determined using principal component analysis. The second test area was the inner
subdivision plot. Here, the Hough Transform was used to emerge the edges. Obtained boundaries,
both for the first and the second test area, were compared with the boundaries from the Polish land
registry database. Performed analyses show that proposed algorithms can define the correct course
of land use boundaries. Analyses were conducted for the purpose of control in the system of direct





 payments for agriculture (Integrated Administration Control System—IACS). The aim of the control
Citation: Borowiec, N.; Marmol, U. is to establish the borders and areas of croplands and to verify the declared group of crops on a given
Using LiDAR System as a Data cadastral parcel. The proposed algorithm—based solely on free LiDAR data—allowed the detection
Source for Agricultural Land of inconsistencies in farmers’ declarations. These mainly concerned areas of field roads that were
Boundaries. Remote Sens. 2022, 14, misclassified by farmers as subsidized land, when in fact they should be excluded from subsidies.
1048. https://doi.org/10.3390/ This is visible in both test areas with areas belonging to field roads with an average width of 1.26 and
rs14041048
3.01 m for test area no. 1 and 1.31, 1.15, 1.88, and 2.36 m for test area no. 2 were wrongly classified as
Academic Editor: Pinliang Dong subsidized by farmers.

Received: 27 October 2021

Keywords: LiDAR system; segmentation; edge detection; agricultural land boundary
Accepted: 17 February 2022
Published: 21 February 2022

Publisher’s Note: MDPI stays neutral

with regard to jurisdictional claims in 1. Introduction
published maps and institutional affil-
Technological progress and development methods of processing spatial data have
iations.
popularized the use and increased the availability of various products presenting infor-
mation about the area. Laser scanning became a dynamically developing technology at
the turn of the 21st century and is finding an increasing number of applications in various
Copyright: © 2022 by the authors.
fields of science. Data from an airborne laser scanning system (ALS) significantly facilitate
Licensee MDPI, Basel, Switzerland. and accelerate the collection of information about the topography and terrain, which leads
This article is an open access article to both cooperation and competition with the technology and products of classical pho-
distributed under the terms and togrammetry and geodesy. Currently, the construction of digital terrain models (DTMs)
conditions of the Creative Commons and digital surface models (DSMs) using LiDAR data is common, but the identification of
Attribution (CC BY) license (https:// agricultural land boundaries based only on point clouds is a complex issue.
creativecommons.org/licenses/by/ Light detection and ranging (LiDAR) is an active remote sensing system that first gen-
4.0/). erates a laser pulse and then records the energy reflected from a given surface. Knowledge

Remote Sens. 2022, 14, 1048. https://doi.org/10.3390/rs14041048 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2022, 14, 1048 2 of 17

of the time of signal generation and the moment of its reception, as well as the properties
of the generated light wave can be used to determine the distance to the object. Airborne
laser scanning, which is performed using a flying plane or helicopter, works according
to this principle. The system uses two main components: a laser scanner, which collects
information about the distance between the scanner and a point on the ground surface, and
a combination of Global Positioning System (GPS) and the inertial navigation system (INS),
whose task is to measure the position and orientation of the system. As a result, data are
acquired in the form of a point cloud [1].
The point cloud is not the final product. It is a set of data (points), defined by spatial
coordinates, with stored information about intensity, RGB color, and echo. This representa-
tion reveals a wealth of information, and when processed into numerical models, it enables
subsequent applications.
Airborne laser scanning and its products in the form of point clouds and digital terrain
models and digital surface models are increasingly used. Precise terrain models contain a
large amount of detailed information. Compared to photogrammetry, they enable the study
of terrain overshadowed by vegetation. Therefore, this study focuses on the possibility of
using LiDAR sensor to identify places of land use changes—agricultural boundaries.
Identification of the course of agricultural boundaries is important to control in the
direct agricultural subsidies system (Integrated Administration Control System—IACS) [2].
The Land Parcel Identification System (LPIS), which is a part of IACS, is a system supporting
direct subsidies to farmers, which depend on the area of crops. Farms with a minimum
agricultural area of at least 1.0 ha, consisting of agricultural parcels of at least 0.1 ha, qualify
for direct payments. The procedure is based on the farmer filling in a declaration, which
involves specifying the area of crops intended for payments. The purpose of the control is
to determine whether the submitted declaration is correct, i.e., whether the land declared
by the farmer is indeed eligible for subsidies. Discrepancies are evidence of irregularities in
the declarations, which should be corrected by the farmers. So far control of applications
within the framework of direct payments for land is carried out by two methods, i.e., field
inspection, most often carried out with the use of GPS technology, and the so-called “photo”
method, based mainly on high-resolution satellite images or aerial images.
This study attempts to answer the question of whether ALS data can be used as a
basis for inspecting agricultural land boundaries. LiDAR data are a powerful source of
spatial information that includes not only coordinates but also intensity, return numbers,
and point cloud classification data. Due to the increasing density of acquired data, ALS is
increasingly used in new fields.
The purpose and novelty of this study was the attempt to automate the detection of
agricultural land edges by using only LiDAR data in the analysis. The innovation of the
method is the use of only airborne laser scanning data to indicate the course of agricultural
land boundaries. Determination of the agricultural land boundary is important in the
process of checking and updating the reference databases of the Land Parcel Identification
System. The proposed algorithm—based only on free LiDAR data—was able to detect
inconsistencies in farmers’ declarations.

2. Literature Review
The data coming from the LiDAR sensor are characterized by high measurement accu-
racy. The resulting digital terrain models and digital surface models depict the surrounding
reality in detail. These features determine the multidirectional use of airborne laser scanning
in various fields of science, such as engineering solutions—calculation 3D displacements of
bridges [3], 3D object detection along the road [4,5], building extraction [6,7], land cover
change detection, and forest succession monitoring [8,9] for heterogeneous land use urban
mapping [10], coastal monitoring [11,12], or archeological research [13,14].
The subject of the LPIS is widely discussed in many publications. Among other
things, researchers compare the LPIS to the national cadastre. Reference [15] evaluated
the extent to which the reference data from the cadastral register are modified in the
Remote Sens. 2022, 14, 1048 3 of 17

LPIS in Poland. Reference [16], using the LPIS of the Republic of Ireland, demonstrated
significant differences in cropland/grassland reporting between an inter-annual based
reporting schema and a land use history approach. Research has also been conducted on
a data model for the collaboration between land administration systems and LPIS [17].
Study [18] focuses on a conceptual model of a large Turkish rural SDI design that combines
the sensor usage and attribute datasets for all types of rural lands. In India, a government
program used high-resolution aerial and satellite orthophotomaps, Global Positioning
System, and electronic total stations (ETSs) to create and update land cadastres in a short
time [19]. Reference [20] also analyzes the quality characteristics of orthoimages for visual
identification of agricultural fields. Reference [21] presents a field boundary detection
technique based on deep learning and a variety of image features which was combined
with the graph-based growing contours (GGCs) method to extract agricultural fields in a
study area in Northern Germany.
Airborne laser scanning is increasingly used in research to identify the type of land
cover [22]. Reference [23] evaluated the use of high-resolution LiDAR for classification
of native and tame grasslands and compared these classifications to the best available
landcover mapping product that is currently available for this area. Reference [24] evaluated
the effectiveness of integrating LiDAR data with high spatial resolution near-infrared digital
imagery for object-based classification of land cover types and dominant tree species, using
decision tree analysis. In [25] the authors proposed a process for objective and automated
identification of agricultural parcel features based on processing and combining Sentinel-2
data (to sense different types of irrigation patterns) and LiDAR data (to detect landscape
elements). Another example of combination cadastral data and remote sensing is in
article [26] where high-resolution multi-spectral WorldView-2 satellite images were used
with the object-oriented approach to image classification and image classification algorithm
creation. The main objective is to compare the results obtained with the traditional methods
of cadastral land evaluation and the results obtained by the methods of remote sensing.
Analyses were performed for an area of Butmir Municipality in Sarajevo. Remote sensing
data such as Sentinel-1 radar images were also used for mapping the different crops in the
Camargue region in Southern France. In this study, deep machine learning was used to
perform land classification [27]. Reference [28] proposed a geographic object-based image
analysis approach to enable semiautomatic land classification and mapping using LiDAR
elevation and intensity data. In [29], the authors used a hybrid capsule network for land
cover classification using multispectral light detection and ranging data. Reference [30]
focused on the extraction of uncultivable trails, ditches, and cultivated field parcels within
farmland on the basis of a LIDAR high-resolution gridded DEM. Reference [31] discussed
the impact that the quality of the digital elevation model has on the final result of landslide
susceptibility modeling. The landslide map was developed on the basis of the analysis of
archival geological maps and the light detection and ranging digital elevation model.

3. Specification of Test Data

The test area is located in Lubelskie Voivodeship in the village of Zimno and includes
only the agricultural lands on which this study focuses. The test area was characterized by
varied terrain. The laser data used in the study were taken from the ISOK project [32]. ISOK
is a Polish information security system created with the aim of improving the protection
of the economy, environment, and society against extraordinary threats, in the first place
against floods. The basic specifications of the ISOK project are: 12 points per square meter,
average distance between the points is 0.3 m, overlap between scans ≥20%, scan cross
angle ≤±25◦ , min scan lane width ≥100 m, and laser beam diameter ≤0.5 m. The obtained
data are in the PL-1992 situation system and PL-KRON86-NH elevation system, with
additional information, i.e., intensity and min 4 echo. In the study area, in accordance with
the Regulation of the Minister of Regional Development and Construction on land and
building cadastres [33], the following agricultural land was distinguished: arable crop land,
marked with the symbol R; permanent grassland, marked by the symbol Ł; and permanent
Remote Sens. 2022, 14, 1048 4 of 17

pastures, marked with the symbol Ps. In the selected area, the agricultural land should
correspond to the cadastral parcels. This would make it possible to check whether there is
any anomalous information in the area, indicating a discrepancy between the declarations
in the payment system. Two study samples were selected for analysis in order to determine
the actual land use status. A visualization of the area is shown in Figure 1.

Figure 1. The figure shows a point cloud acquired from the airborne laser scanning system. The
cloud is displayed in plan view. (a) displays natural RGB colors, (b) the displayed point cloud uses
the intensity, (c) the color corresponds to the height of the point, that is the Z coordinates.

4. Scheme of the Proposed Algorithm

The proposed algorithm consists of two stages. In the first stage, a point cloud is
interpolated into a regular grid with a resolution of 0.3 m. Nearest neighbor interpolation
was used. Two algorithms are used with the obtained rasters: edge detection and segmen-
tation, on the basis of which the approximate location of agricultural land boundaries is
estimated. In the next stage, we returned to the raw point cloud for pre-selected areas from
the first part. Two methods were used for the original data: PCA and Hough Transform,
which allowed for precise determination of agricultural land boundaries. The scheme of
the algorithm is shown in Figure 2.

Figure 2. The scheme of the proposed algorithm.

5. Detection of Agricultural Land Boundaries from Raster Data

As a first step, an initial detection of agricultural land was undertaken using a point
cloud stored as a raster. For this purpose, rasters with a resolution of 0.3 m were generated—
images of intensity, classification, height differences, and RGB values.
Remote Sens. 2022, 14, 1048 5 of 17

An edge detection and segmentation process were performed in the MATLAB envi-
ronment.

5.1. Edge Detection

Edge detection was carried out using the combined operators of Prewitt and Canny [34],
based on intensity and classification rasters. It was decided to combine the two methods
in order to obtain a more reliable agricultural land boundary. The Prewitt operator, based
on gradients, more accurately determined the edges, but the information about them was
point-wise. The Canny operator, on the other hand, introduced linear information but detected
undesirable ploughing traces in the fields. Ploughing traces were recorded as lines, which
made it difficult to eliminate them in subsequent steps. In the next step, simple morphological
operators (erosion and dilation) were applied to remove noise in the image. The performed
work made it possible to eliminate many errors and better illustrate land use boundaries in
the analyzed area. However, errors are still visible. The result of this stage is presented in
Figure 3.

Figure 3. Edge detection of agricultural lands along with the visible errors—noise. The selected noise
is marked with red circles.

5.2. Segmentation
Segmentation was used as a further supporting step. The aim was to roughly de-
termine the areas of individual agricultural lands. Segmentation was carried out on the
intensity image.
It was decided to use the multi-resolution algorithm (MRS). This is one of the most
widely used segmentation models. It is based on the minimalization of the average hetero-
geneity in a single object extracted from an image [35]. This algorithm was chosen because a
literature review suggested that this segmentation method would be more accurate [36,37].
The first key parameter of the algorithm is scale, i.e., the size of objects (segments) in
the analyzed area. In this study, the value of 200 is used. In addition to the scale parameter,
two additional parameters are required: shape and compactness [38]. The best results
were obtained for values of 0.9 and 0.1, respectively. The segmentation result is shown in
Figure 4.
The performed work allowed for a preliminary identification of the areas of agricul-
tural fields, as well as their boundaries. The effect of the work is illustrated in Figure 5.
As can be seen from Figure 5, noise—incorrect edges in object detection—is still
present. The integration of edges and segments can be used to identify some of them.
Incorrect edges, related to ploughing traces in the fields, were located by correct segments
in this area—this mainly concerns three parcels in the central-western part of the study
and one parcel in the south-eastern part. Furthermore, in a few parts the segmentation did
not work properly. This was due to the similarity of intensities in the neighboring parcels
(see Figure 1b). The different edges in these areas indicate the locations of possible errors
(mainly the central and central-southern part of the study). The stage described in this
Remote Sens. 2022, 14, 1048 6 of 17

subsection is considered as a prelude to further work aimed at more accurate identification

of agricultural land boundaries.

Figure 4. Segmentation of agricultural lands in the study area.

Figure 5. Overlaying of the segmentation and edge detection steps.

6. Accurate Detection of Agricultural Land Boundaries

After identification of initial regions of agricultural land boundaries, there was a return
to the original point cloud, but this was limited to the indicated sub-areas defining the
course of agricultural land boundaries. It was decided to return to distributed data because
there are several airborne laser scanning points per image pixel, which will significantly
increase the accuracy of the edge location (Figure 6). Therefore, the original data increase
the amount of information that can define the agricultural land with higher accuracy.

Figure 6. Scattered points superimposed on a raster representing parcel boundaries.

In this study, precise detection of agricultural boundaries was presented for two test
areas (segments). In the first area, the segment boundaries coincided entirely with a single
Remote Sens. 2022, 14, 1048 7 of 17

area of agricultural land. In contrast, the second selected segment included additional
agricultural boundaries within its area. Therefore, two approaches for precise identification
of agricultural boundaries with the use of scattered data (irregular point cloud) were
applied in this study. In the first one, a solution based on principal component analysis
was used to determine land parcel boundaries. On the other hand, the second approach
used Hough Transform.

6.1. Boundary Detection Using Principal Component Analysis—Test Area No. 1

The first test field is an area of agricultural use whose extent overlaps with the rough
edges delineated in stage 1 (Figure 7).

Figure 7. Test area no. 1—segment, coinciding with approximate edges.

Analyses began by indicating the test area (segment) from the segmentation image
(Figure 8a), then a product was performed with the raster depicting the edges of the
agricultural (Figure 8b), resulting in an image showing only the edges of the segment
(Figure 8c).

Figure 8. The sequence steps (a–c) leading to a raster edge representation of one segment.

The next stage leading to a correct determination of the edges was grouping the set of
points into four parts containing points describing each parcel edge separately. For this
purpose, a solution based on principal component analysis (PCA) was used. This solution
allows us to determine the scatter model of the dataset in the form of a probability ellipse.
Each pixel in the image is defined by two variables, the X variable and the Y variable
(column number, row number). The ellipse is plotted based on the assumption that the
given two variables follow a two-dimensional normal distribution (Gaussian distribution).
The orientation of ellipse depends on the sign of correlation coefficient between variables,
the size of the ellipse is determined by the intervals and its center is defined by the averages
of the variables X and Y. The term intervals refers to the root of the eigenvalues multiplied
by the user-selected value. Eigenvalues can be interpreted as proportions of the variance
explained by correlations between relevant variables. Therefore, in solving this task, the
values of variance and covariance were calculated for the variables X and Y to build the
designated perimeter of the object.
The value of variance, which determines the diversity of the community, is equal to
the sum of the arithmetic mean of squares of deviations of individual feature values from
Remote Sens. 2022, 14, 1048 8 of 17

the arithmetic mean of the community. The unconstrained variance estimator for X and Y
coordinates was calculated from the following formulas [39]:

n 2 n 2
∑ ( x − xi ) ∑ ( y − yi )
i =1 i =1
sx 2 = sy 2 = , (1)
n−1 n−1

S2x /Sy2 —the variance for X/Y calculated from the sample—unbiased variance estima-
tor;
cov(X,Y)—covariance of a variables X, Y set;
x/y—sample mean value for X/Y;
xi , yi /—value of the X/Y variable for the i-th point
n—sample size.
The covariance value, which determines the linear relationship between the random
variables X and Y, can be calculated from the following formula:

1 n
n i∑
cov( X, Y ) = ( x − x i ) · ( y − y i ), (2)
=1

As a result of the calculations performed, the variance–covariance matrix can be

S2x
 
cov( X, Y )
constructed for X, Y coordinates. From the obtained matrix, the
cov( X, Y ) Sy2
eigenvalues of the matrix can be calculated.
The obtained variance–covariance matrix and eigenvalues were used to calculate
the ellipse parameters, i.e., orientation and size of the ellipse according to Hausbrandt’s
formulas [40].  
2·cov( X,Y )
arctan s2x −s2y
Ω= , (3)
2
Ω—the omega angle between the horizontal line and the direction of the eigenvector
with the larger eigenvalue;
S2x /Sy2 —the variance for X/Y calculated from the sample—unbiased variance estima-
tor;
cov(X,Y)—covariance of a variables X, Y set.
The lengths of the minor (a) and major (b) half-axes of the confidence area ellipse were
calculated from the formulas below:
√ √
e1 = 2 · a1 , e2 = 2 · a2 , (4)

e1 —the length value for the minor half-axis of the ellipse;

e2 —the length value for the major half-axis of the ellipse;
a1 —minor eigenvalue of the test area object;
a2 —major eigenvalue of the test area object.
In the above formula, the square root of the eigenvalues was multiplied by 2, which
means that 95.5% of the feature values lie at a distance ≤2 from the expected value. The
center of gravity was calculated as the arithmetic mean of the X, Y coordinates, which
determined the midpoint of the ellipse. The obtained parameters can be used to determine
the span of the probability ellipse and to determine the main directions along which the
points defining the boundaries of the parcel are arranged (Figure 9).
Remote Sens. 2022, 14, 1048 9 of 17

Figure 9. Example of an error ellipse calculated for one of the parcels of land.

The parameters of the ellipse also made it possible to divide the points into four sets of
data that define four corresponding boundaries. In each of the four sets of points, a simple
least squares approximation was performed (Figure 10). The approximation is an iterative
process that, in successive iterations, discards outlier points from the straight line until
the agricultural edges are accurately determined. Outlier points are those points that lie
further than the average distance of the points from the straight line obtained in successive
iterations. The iterative process ends when the sum of the squares of the outliers reaches a
minimum.

Figure 10. Approximation of straight lines based on scattered points that define utility boundaries.

The intersections of the detected lines determined the vertices of the sought parcel.

6.2. Boundary Detection Using Hough Transform—Test Area No. 2

The second test field was an erroneously delineated segment. The rough edges
produced in the first stage indicate that there is probably more agricultural land here
(Figure 11).

Figure 11. Test area no. 2—segment with more approximate edges.

Similar to test area no. 1, a segment was first selected on which a raster representing
the edges of the utilities was overlaid, resulting in a binary image containing only the edges
of the segment (Figure 12a–c).
Remote Sens. 2022, 14, 1048 10 of 17

Figure 12. The next steps (a–c) detected the outer and inner edges of the selected segment.

The selected segment contains not only edges describing its perimeter but also several
edges inside the area. The study showed that the principal components method cannot
automatically write to separate sets of points representing each edge (those inside and
outside). Therefore, Hough Transform was used to select and identify all the edges of the
analyzed area. Hough Transform enables fast detection of straights in a binary image [41].
When detecting collinear pixels in an image in this method, it is possible to indicate the
number of straights that need to be detected. In this study, nine lines were detected
(Figure 13).

Figure 13. A binary raster representing the edges with the detected lines by Hough Transform.

Based on the detected lines, it was possible to group the scattered points into appro-
priate sets, defining the course of the edge. The condition determining points belonging to
a given edge was the distance of the point from the straight line. Thus, nine sets of points
were defined, and in each set a straight line was approximated by the least squares method
(Figure 14).

Figure 14. Approximation of straight lines defining the course of agricultural edges.

In the next step, the vertices of the areas where the lines intersect were determined.
Next, the vector cadastral data were overlaid on the raster representing the edges of the
agricultural land (Figure 15). According to IACS, there can be several types of agricultural
land in each parcel of land. However, after visual verification it was found that in the third
segment there is one agricultural area of land in the whole cadastral parcel. Therefore, the
inclusion of two straights in further analyses was discontinued. The straights are indicated
by arrows in Figure 15.
Remote Sens. 2022, 14, 1048 11 of 17

Figure 15. Land use boundaries and overlaid vectors from land records. Arrows indicate boundaries
that were not considered in further analyses.

7. Analysis and Discussion of the Obtained Results

7.1. Test Area No. 1
Test area no. 1 was the arable crop land (R) corresponding to cadastral parcel no. 700.
The analysis of the obtained results consisted of comparing the borders of the agricultural
land obtained by means of the PCA (Section 6.1) with the borders of the agricultural land
from the cadastral records. For this purpose, additional points were inserted on both lines
at corresponding intervals of 0.5 m. Next, deviations were determined as distances between
corresponding points.
Western boundary (W): Conformity of agricultural land boundaries with the data
recorded in the land cadastre was observed. Obtained deviations are within the range from
0.23 to 0.35 m. The average value is 0.29 m.
Southern boundary (S): Much larger differences are obtained. Deviations range from
0.37 to 2.16 m, with an average value of 1.26 m. Anomalous information is observed,
showing a discrepancy between the declaration and the actual state of use (Figure 16a).
There is a road running in the area, and its boundary is not clear, resulting in changes in
the actual state of use for these parts of the parcel.

Figure 16. Comparison of the southern boundary of the land parcel (S) (a) (red line—boundary from
the land cadastre, blue—from the LiDAR) and the northern boundary (N) (b) (red line—boundary
from the land cadastre, blue—from the LiDAR data).

Eastern boundary (E): Conformity of agricultural land boundaries with the data
recorded in the land cadastre was observed. Deviations range from 0.13 to 0.39 m. The
average value is 0.26 m.
Northern boundary (N): The largest differences were obtained for the northern bound-
ary of the agricultural land (Figure 16b). LiDAR determined a completely different course
of the agricultural land boundary than indicated by the land cadastre. Deviations range
from 1.81 to 4.21 m. The average value of deviations is as high as 3.01 m. Analyzing the data
Remote Sens. 2022, 14, 1048 12 of 17

against the intensity map, it can be observed that both the LiDAR data and the cadastre
data show discrepancies in relation to the actual land use.
The analysis is presented in two tables. Table 1 gives the summarized results for the
analyzed agricultural land.

Table 1. Basic statistics for agricultural land R (parcel no. 700).

Agricultural Minimum Maximum Average

Parcel Boundary Remarks
Land Deviation (m) Deviation (m) Deviation (m)
conformity of the
boundary W 0.23 0.35 0.29 declaration with the
actual state
non-conformity of the
boundary S 0.37 2.16 1.26 declaration with the
arable crop actual state—field road
land—R 700
conformity of the
boundary E 0.13 0.39 0.26 declaration with the
actual state
non-conformity of the
boundary N 1.81 4.21 3.01 declaration with the
actual state—field road

It is noted that the largest amount of anomalous information occurs for areas of
agricultural land boundaries with field roads. Table 2 is a comparison carried out for
two variants: The first considers the situation of the agricultural land–agricultural land
boundary and the second the agricultural land–field road.

Table 2. Anomalous information for variant 1, land–land boundary, and variant 2, land–field road
boundary.

Minimum Maximum Average Deviation

Boundary
Deviation (m) Deviation (m) (m)
land–land boundary 0.13 0.39 0.27
land–field road
0.37 4.21 2.01
boundary

7.2. Test Area No. 2

After the analysis of the cadastral data, it was revealed that the selected segment
consists of three agricultural lands—R, Ł, corresponding to cadastral parcels with the
numbers: 702 (R), 705 (R), and 707 (Ł). The algorithm based on Hough Transform correctly
extracted these three areas.
As a first step, a detailed analysis of the agricultural land R for parcel 702 was carried
out. Western boundary (W): Conformity of agricultural land boundaries with the data
recorded in the land cadastre was observed. Deviations range from 0.23 to 0.40 m. The
average value is 0.31 m.
Southern boundary (S): Much larger differences were obtained. The deviations range
from 1 to 1.61 m, with an average deviation of 1.31 m. In this area we are dealing with a
field road, the course of which is unclear. A discrepancy between the declaration and the
actual state of use can be seen in the form of a systematic shift between the LiDAR data
and the agricultural land from the land cadastre (Figure 17a).
Remote Sens. 2022, 14, 1048 13 of 17

Figure 17. Comparison of the southern boundary of the land parcel (S) (a) (red line—boundary from
the land cadastre, blue—from the LiDAR) and the northern boundary (N) (b) (red line—boundary
from the land cadastre, blue—from the LiDAR).

Eastern boundary (S): For the eastern boundary, deviations of 0.40–0.70 m were ob-
tained, with a mean value of 0.55 m.
Northern boundary (N): Much larger differences were obtained (Figure 17b). The
deviations range from 0.66 to 1.96 m, and the mean value of the deviations is 1.15 m. In this
case a field road is also located in the area.
Due to the repeated spatial situation, the results for parcels 705 and 707 are grouped
together in Table 3. For both parcels similar types of differences to parcel 702 appeared, as
discussed above. The results for plot 702 are also included in the table.

Table 3. Basic statistics for parcels 702, 705, and 707.

Agricultural Minimum Maximum Average

Parcel Boundary Remarks
Land Deviation (m) Deviation (m) Deviation (m)
conformity of the
boundary W 0.23 0.40 0.31 declaration with the
actual state
non-conformity of the
boundary S 1.0 1.61 1.31 declaration with the
arable crop actual state—field road
land—R 702
conformity of the
boundary E 0.40 0.70 0.55 declaration with the
actual state
non-conformity of the
boundary N 0.66 1.96 1.15 declaration with the
actual state—field road
conformity of the
boundary W 0.40 0.70 0.55 declaration with the
actual state
non-conformity of the
boundary S 1.72 2.33 1.88 declaration with the
arable crop actual state—field road
land—R 705
conformity of the
boundary E 0.55 1.22 0.88 declaration with the
reality state
conformity of the
boundary N 0.72 0.77 0.74 declaration with the
actual state—field road
Remote Sens. 2022, 14, 1048 14 of 17

Table 3. Cont.

Agricultural Minimum Maximum Average

Parcel Boundary Remarks
Land Deviation (m) Deviation (m) Deviation (m)
conformity of the
boundary W 0.55 1.22 0.88 declaration with the
reality state
non-conformity of the
boundary S 1.86 2.87 2.36 declaration with the
actual state—field road
grassland—Ł 707
conformity of the
boundary E 0.00 0.42 0.16 declaration with the
reality state
conformity of the
boundary N 0.31 0.64 0.47 declaration with the
actual state—field road

A similar rule was observed as in the previous analyses, i.e., a much higher agreement
of boundaries for the land–land boundary variant than for the land–field road boundary
variant. This is presented in Table 4.

Table 4. Anomalous information for variant 1, land–land boundary, and variant 2, land–field road
boundary.

Minimum Maximum Average

Boundary
Deviation (m) Deviation (m) Deviation (m)
land–land boundary 0.00 1.22 0.47
land–field road
0.31 2.87 1.31
boundary

8. Conclusions
In summary, the proposed algorithms make it possible to carry out controls in the
system of direct payments to agriculture. The use of laser data makes it possible to
determine specific agricultural land and to determine the size of anomalous information.
Correctness has been noted in relation to the conformity of the actual statue with the
declaration in the case of boundaries of adjacent agricultural land (variant 1: land—land).
However, in areas where there are borders of agricultural land with field roads, there are
visible discrepancies (variant 2: land—field road). In these areas, anomalous information
connected with the difference between the actual use of a given area and the legal status
recorded in the declaration should be noted.
The obtained results were considered satisfactory. LiDAR proved to be very useful
technology in the process of detecting agricultural boundaries. Most of the boundaries
were readable in the laser data.
The advantage of using a point cloud over traditional aerial images is that an additional
elevation information can be used. In the case of aerial images, analyses are carried out on
2D data stored as a raster. In the developed algorithm, the second step returns to the raw
point cloud. The knowledge of an additional Z coordinate may allow for more precise edge
detection in areas where the 2D information is ambiguous.
It was noted that the introduction of higher resolution data would certainly contribute
to an increase in accuracy. The use of laser data from a UAV flight would allow a more
precise determination of boundaries in doubtful cases, not fully legible in the case of the
data used in this study (average distance between points 0.3 m).
The two-stage approach to analysis also proved to be a valuable solution. Edge
detection and segmentation algorithms used in the first stage allowed us to roughly estimate
Remote Sens. 2022, 14, 1048 15 of 17

the area and boundaries of individual agricultural lands. In the second stage, we returned to
the original data for the locations presenting the contours of the agriculture boundaries. In
the developed method of detecting straights, two approaches were used. Two approaches
were chosen because the first area is a rectangular parcel of land and the second test
area is a rectangle with an internal land boundary, and precise determination of the land
boundary is possible if only one boundary is displayed in the raster image. Therefore,
in each approach, the aim was to divide the point cloud into datasets representing only
one land use boundary. With such datasets it is possible to approximate straights with
higher accuracy than by using raster data. Detected land use boundaries are described by
the equation of the straight line, determined in an iterative process, wherein subsequent
iterations’ outliers are rejected. Thus, the obtained straight line reliably reflects the course
of land use boundaries detected based on ALS data.
The created algorithm allowed the detection of inconsistencies in farmers’ declarations.
These were related to areas of field roads that were incorrectly declared by farmers as
donated land, when in fact they should be excluded from subsidies. It was detected that
both test areas, test field 1, areas belonging to field roads with an average width of 1.26 and
3.01 m, and test field 2, areas belonging to field roads with an average width of 1.31, 1.15,
1.88, and 2.36 m, were incorrectly classified by farmers as donated land.
In this study the authors focus on identification of land boundaries between land
uses covered with low vegetation. In the next research, the authors intend to analyze the
boundaries of parcels that are farmed and covered with different species of plants. Such
diversity can help at the stage of segmentation because each species can have a different
intensity value. Different intensity value will contribute to easier initial identification of
parcel edges. At the stage of precise identification of plot boundaries, on the other hand, it
may be necessary to select only points reflected from the ground. Then it will be necessary
to perform a filtering of the lidar data to be used for further analysis.
In conclusion, the method used in this study based on LiDAR data is useful for
automatic verification and monitoring of anomalous information showing inconsistency of
the declaration with the actual state of land use.

Author Contributions: Conceptualization, N.B. and U.M.; methodology, N.B. and U.M.; software,
N.B. and U.M.; validation N.B. and U.M.; formal analysis N.B. and U.M.; writing—original draft
preparation, N.B. and U.M.; writing—review and editing, N.B. and U.M.; visualization, N.B. and
U.M.; supervision, N.B. and U.M.; project administration, N.B. and U.M. All authors have read and
agreed to the published version of the manuscript.
Funding: Research project supported by program „Excellence initiative—research university” for the
AGH University of Science and Technology no. 501.696.7996. The project title: Integration of remote
sensing data for control in the system of direct agricultural subsidies (IACS).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study (.las files) are available in ISOK
project—https://isok.gov.pl/index.html, accessed on 31 July 2021.
Acknowledgments: The authors thank the editor and anonymous reviewers for their helpful com-
ments and valuable suggestions.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Vosselman, G.; Mass, H.-G. Airborne and Terrestrial Laser Scanning; Whittles Publishing: Dunbeath, UK, 2010.
2. Prawa i Obowiazki˛ Rolników w Procesie Kontroli na Miejscu|Agencja Restrukturyzacji i Modernizacji Rolnictwa. Available
online: https://www.arimr.gov.pl/kontrole-beneficjentow/prawa-i-obowiazki-rolnikow-w-procesie-kontroli-na-miejscu.html
(accessed on 22 August 2021).
3. Cha, G.; Sim, S.-H.; Park, S.; Oh, T. LiDAR-Based Bridge Displacement Estimation Using 3D Spatial Optimization. Sensors 2020,
20, 7117. [CrossRef] [PubMed]
Remote Sens. 2022, 14, 1048 16 of 17

4. Chen, Z.; Lin, Q.; Sun, J.; Feng, Y.; Liu, S.; Liu, Q.; Ji, Y.; Xu, H. Cascaded Cross-Modality Fusion Network for 3D Object Detection.
Sensors 2020, 20, 7243. [CrossRef] [PubMed]
5. Imad, M.; Doukhi, O.; Lee, D.-J. Transfer Learning Based Semantic Segmentation for 3D Object Detection from Point Cloud.
Sensors 2021, 21, 3964. [CrossRef] [PubMed]
6. Dey, E.K.; Tarsha Kurdi, F.; Awrangjeb, M.; Stantic, B. Effective Selection of Variable Point Neighbourhood for Feature Point
Extraction from Aerial Building Point Cloud Data. Remote Sens. 2021, 13, 1520. [CrossRef]
7. Huang, J.; Zhang, X.; Xin, Q.; Sun, Y.; Zhang, P. Automatic building extraction from high-resolution aerial images and LiDAR
data using gated residual refinement network. ISPRS J. Photogramm. Remote Sens. 2019, 151, 91–105. [CrossRef]
8. Szostak, M. Automated Land Cover Change Detection and Forest Succession Monitoring Using LiDAR Point Clouds and GIS
Analyses. Geosciences 2020, 10, 321. [CrossRef]
9. Gu, C.; Zhai, C.; Wang, X.; Wang, S. Cmpc: An innovative lidar-based method to estimate tree canopy meshing-profile volumes
for orchard target-oriented spray. Sensors 2021, 21, 4252. [CrossRef]
10. Megahed, Y.; Shaker, A.; Yan, W.Y. Fusion of Airborne LiDAR Point Clouds and Aerial Images for Heterogeneous Land-Use
Urban Mapping. Remote Sens. 2021, 13, 814. [CrossRef]
11. Terefenko, P.; Zelaya Wziatek,˛ D.; Dalyot, S.; Boski, T.; Pinheiro Lima-Filho, F. A High-Precision LiDAR-Based Method for
Surveying and Classifying Coastal Notches. ISPRS Int. J. Geo-Inf. 2018, 7, 295. [CrossRef]
12. Swirad, Z.M.; Young, A.P. Automating coastal cliff erosion measurements from large-area LiDAR datasets in California, USA.
Geomorphology 2021, 389, 107799. [CrossRef]
13. Vilbig, J.M.; Sagan, V.; Bodine, C. Archaeological surveying with airborne LiDAR and UAV photogrammetry: A comparative
analysis at Cahokia Mounds. J. Archaeol. Sci. Rep. 2020, 33, 102509. [CrossRef]
14. Balsi, M.; Esposito, S.; Fallavollita, P.; Melis, M.G.; Milanese, M. Preliminary Archeological Site Survey by UAV-Borne Lidar: A
Case Study. Remote Sens. 2021, 13, 332. [CrossRef]
15. Kocur-Bera, K. Understanding information about agricultural land. An evaluation of the extent of data modification in the Land
Parcel Identification System for the needs of area-based payments—A case study. Land Use Policy 2020, 94, 104527. [CrossRef]
16. Zimmermann, J.; González, A.; Jones, M.B.; O’Brien, P.; Stout, J.C.; Green, S. Assessing land-use history for reporting on cropland
dynamics-A comparison between the Land-Parcel Identification System and traditional inter-annual approaches. Land Use Policy
2016, 52, 30–40. [CrossRef]
17. Inan, H.I.; Sagris, V.; Devos, W.; Milenov, P.; van Oosterom, P.; Zevenbergen, J. Data model for the collaboration between land
administration systems and agricultural land parcel identification systems. J. Environ. Manag. 2010, 91, 2440–2454. [CrossRef]
18. Iban, M.C.; Aksu, O. A model for big spatial rural data infrastructure in Turkey: Sensor-driven and integrative approach. Land
Use Policy 2020, 91, 104376. [CrossRef]
19. Parida, P.K.; Sanabada, M.K.; Tripathi, S. The Digital Cadastral Map/Layer Generation and Conclusive Titling of Land Parcels
Using Hybrid Technology (Aerial/High-Resolution Image (HRSI) and DGPS and ETS Survey) Adopted by Govt. of Odisha
Under Digital India Land Record Modernization Programme (DIL). In Lecture Notes in Civil Engineering; Springer: Singapore,
2020; Volume 33, pp. 439–459.
20. Tarko, A.; de Bruin, S.; Fasbender, D.; Devos, W.; Bregt, A.K. Users’ assessment of orthoimage photometric quality for visual
interpretation of agricultural fields. Remote Sens. 2015, 7, 4919–4936. [CrossRef]
21. Wagner, M.P.; Oppelt, N. Deep Learning and Adaptive Graph-Based Growing Contours for Agricultural Field Extraction. Remote
Sens. 2020, 12, 1990. [CrossRef]
22. Kamal, M.; Phinn, S.; Johansen, K. Object-based approach for multi-scale mangrove composition mapping using multi-resolution
image datasets. Remote Sens. 2015, 7, 4753–4783. [CrossRef]
23. Fisher, R.J.; Sawa, B.; Prieto, B. A novel technique using LiDAR to identify native-dominated and tame-dominated grasslands in
Canada. Remote Sens. Environ. 2018, 218, 201–206. [CrossRef]
24. Sasaki, T.; Imanishi, J.; Ioki, K.; Morimoto, Y.; Kitada, K. Object-based classification of land cover and tree species by integrating
airborne LiDAR and high spatial resolution imagery data. Landsc. Ecol. Eng. 2012, 8, 157–171. [CrossRef]
25. Estrada, J.; Sánchez, H.; Hernanz, L.; Checa, M.; Roman, D. Enabling the Use of Sentinel-2 and LiDAR Data for Common
Agriculture Policy Funds Assignment. ISPRS Int. J. Geo-Inf. 2017, 6, 255. [CrossRef]
26. Mulahusić, A.; Topoljak, J.; Tuno, N.; Ajvazović, K. The Possibilities of the Cadastral Land Use Assessment by the Methods of
Remote Sensing. In Lecture Notes in Networks and Systems; Springer Nature: Cham, Switzerland, 2019; Volume 42, pp. 452–458.
27. Ho Tong Minh, D.; Ndikumana, E.; Baghdadi, N.; Courault, D.; Hossard, L. Applying deep learning for agricultural classification
using multitemporal SAR Sentinel-1 for Camargue, France. In Proceedings of the Image and Signal Processing for Remote Sensing
XXIV; Bruzzone, L., Bovolo, F., Benediktsson, J.A., Eds.; SPIE: Berlin, Germany, 2018; Volume 10789, p. 39.
28. Zhou, X.; Li, W. A Geographic Object-Based Approach for Land Classification Using LiDAR Elevation and Intensity. IEEE Geosci.
Remote Sens. Lett. 2017, 14, 669–673. [CrossRef]
29. Yu, Y.; Guan, H.; Li, D.; Gu, T.; Wang, L.; Ma, L.; Li, J. A Hybrid Capsule Network for Land Cover Classification Using
Multispectral LiDAR Data. IEEE Geosci. Remote Sens. Lett. 2020, 17, 1263–1267. [CrossRef]
30. Qian, T.; Shen, D.; Xi, C.; Chen, J.; Wang, J. Extracting Farmland Features from Lidar-Derived DEM for Improving Flood Plain
Delineation. Water 2018, 10, 252. [CrossRef]
Remote Sens. 2022, 14, 1048 17 of 17

31. Kamiński, M. The Impact of Quality of Digital Elevation Models on the Result of Landslide Susceptibility Modeling Using the
Method of Weights of Evidence. Geosciences 2020, 10, 488. [CrossRef]
32. ISOK. Available online: https://isok.gov.pl/index.html (accessed on 31 July 2021).
33. Regulation of the Minister for Regional Development and Construction of 29 March 2001 on the land and building cadastre.
Journal of Laws, No. 38 item 454 as Amended. 2001.
34. Gonzalez, R.C.; Woods, R.E. Digital Image Processing, 3rd ed.; Pearson: Upper Saddle River, NJ, USA, 2008; ISBN 978-0-13-168728-8.
35. Baatz, M.; Schape, A. Multiresolution segmentation—An optimization approach for high quality multi-scale image segmentation.
In Proceedings of the AGIT Symposium, Salzburg, Austria, 5–7 July 2000.
36. Kavzoglu, T.; Tonbul, H. A comparative study of segmentation quality for multi-resolution segmentation and watershed transform.
In Proceedings of the 2017 8th International Conference on Recent Advances in Space Technologies RAST 2017, Istanbul, Turkey,
19–22 June 2017; pp. 113–117. [CrossRef]
37. Kavzoglu, T.; Tonbul, H. An experimental comparison of multi-resolution segmentation, SLIC and K-means clustering for
object-based classification of VHR imagery. Int. J. Remote Sens. 2018, 39, 6020–6036. [CrossRef]
38. Benz, U.C.; Hofmann, P.; Willhauck, G.; Lingenfelder, I.; Heynen, M. Multi-resolution, object-oriented fuzzy analysis of remote
sensing data for GIS-ready information. ISPRS J. Photogramm. Remote Sens. 2004, 58, 239–258. [CrossRef]
39. Sobczyk, M. Statystyka; Wydawnictwo Naukowe PWN: Warsaw, Poland, 2007.
40. Hausbrandt, S. Rachunek Wyrównawczy i Obliczenia Geodezyjne; Państwowe Przedsi˛ebiorstwo Wydawnictw Kartograficznych:
Warsaw, Poland, 1970.
41. Marmol, U.; Borowiec, N. Detection of Line Objects by Means of Gabor Wavelets and Hough Transform. Arch. Civ. Eng. 2020, 66,
339–363. [CrossRef]