IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 1, March 2024, pp. 1129~1138

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i1.pp1129-1138  1129

Backbone search for object detection for applications in

intrusion warning systems

Nguyen Duc Thuan, Nguyen Thi Lan Huong, Hoang Si Hong

School of Electrical and Electronic Engineering, Hanoi University of Science and Technology, Hanoi, Vietnam

Article Info ABSTRACT

Article history: In this work, we propose a novel backbone search method for object
detection for applications in intrusion warning systems. The goal is to find a
Received Apr 8, 2023 compact model for use in embedded thermal imaging cameras widely used
Revised Oct 15, 2023 in intrusion warning systems. The proposed method is based on faster
Accepted Nov 9, 2023 region-based convolutional neural network (Faster R-CNN) because it can
detect small objects. Inspired by EfficientNet, the sought-after backbone
architecture is obtained by finding the most suitable width scale for the base
Keywords: backbone (ResNet50). The evaluation metrics are mean average precision
(mAP), number of parameters, and number of multiply–accumulate
Architecture search operations (MACs). The experimental results showed that the proposed
Dataset method is effective in building a lightweight neural network for the task of
Neural network object detection. The obtained model can keep the predefined mAP while
Object detection minimizing the number of parameters and computational resources. All
Thermal image experiments are executed elaborately on the person detection in intrusion
warning systems (PDIWS) dataset.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Hoang Si Hong
School of Electrical and Electronic Engineering, Hanoi University of Science and Technology
No. 1 Dai Co Viet Road, Hanoi, Vietnam
Email: [email protected]

1. INTRODUCTION
In recent years, object detection algorithms using deep learning have achieved remarkable results.
They have been widely used in the implementation of practical systems such as intrusion warning systems
[1], [2]. In these systems, object detection algorithms are used to detect human intrusions early by processing
images from surveillance cameras [3]. With large systems, processing locally is essential because it is timely
and reduces the load on the central server [4]. However, implementing deep learning algorithms on such
device surveillance cameras is generally difficult because of hardware constraints [5]. Therefore, scientists
have proposed methods to compact deep learning models, to efficiently implement them in practice.
In general, there are two approaches to deep learning model compaction: network architecture
search (NAS) and model compression [6]. For the first approach, an optimal network architecture will be
searched in a large space of potential architectures. This method is proven to be effective in cases where the
designer has almost no knowledge of the data used [7]. The popular methods for object detection problems
can be mentioned as: Detection NAS (DetNAS) [8], method in [9], feature pyramid network NAS (FPN-
NAS) [10], auto feature pyramid network (Auto-FPN) [11], structural-to-modular NAS (SM-NAS) [12]. The
common property of these methods is that they give a relatively good performance because the preset criteria
are achieved during searching. In return, the computational cost for them including training time, power
consumption is often very expensive and often have to run on powerful hardware.

Journal homepage: http://ijai.iaescore.com

1130  ISSN: 2252-8938

For the second approach, the design of the neural network requires the programmer to have data
knowledge such as the variety of data based on previous studies [13]. After that, a regular neural network
model called the base model will be established without regard to the compaction criteria. To optimize the
network, one will take advantage of model compression methods including pruning and quantization. These
methods observe the base model to remove unnecessary parameters or approximate parameters with fewer
byte representations. Methods of this type are usually not specific to any task, but rather general in machine
learning. Well-known methods can be mentioned as: Filter pruning via geometric median (FPGM) [14],
method in [15], method in [16], quantization after training (QAT) [17] and binarized neural networks (BNN)
[18]. Compared with neural network architecture search, model compression methods significantly reduce
the computational cost while maintaining the optimal criteria.
In this work, to save computational resources, we propose a simple neural network architecture
search model for object detection. We refer to previous literature to design an object detection model based
on faster region-based convolutional neural network (Faster R-CNN) [19] and ResNet50 [20]. We were
inspired by EfficientNet [21] to find the width scale for the base model ResNet50 so that the scaled model
meets the criteria for mAP, the number of parameters and the number of multiply-accumulate operation
(MAC) operations. The reason we only looked for the width scale is that we found that many efficient object
detection models use the feature pyramid network (FPN) structure [22], so scaling to the width should not
affect it. Our main motivation is to suggest a simple, low-cost method that still guarantees reservation
requirements.

2. RELATED WORKS
2.1. Faster R-CNN
The premier model within the R-CNN family is Faster R-CNN, introduced in 2015 [19]. Subsequent
versions of R-CNN family networks have undergone enhancements primarily focused on computational
efficiency, incorporating diverse training phases, reducing inference time, and boosting overall performance
measured by mean average precision (mAP). These networks typically consist of: a) an algorithm to find
"bounding boxes" or possible object positions in the image; b) the stage of extracting the features of the
object, usually using a convolution neural network (CNN) network; c) a classification network to predict the
class of object; and d) a regression layer to make the coordinates of the bounding boxes more accurate.
Faster R-CNN combines 2 modules shown in Figure 1. The first module is to use a deep neural
network (DNN) to propose the regions, called region proposal network (RPN) and the second module is the
Fast R-CNN model that uses the proposed regions. Fast R-CNN will take the suggested regions from the
RPN to determine the object corresponding to the anchor. Faster R-CNN also has a CNN backbone network,
a region of interest (ROI) pooling layer and a full connectivity layer, followed by two sub-branches to
perform the two tasks of classifying objects and finding the best bounding box based on the regression.

Figure 1. Faster R-CNN architecture [23]

2.1.1. Region proposal network (RPN)

Region proposal network RPN is the main innovation that makes faster R-CNN the best in the R-
CNN family. RPN solves problems by training the neural network to substitute for selective search in
previous versions, which is very slow. A region proposal network takes as input an image of any size and

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1129-1138

Int J Artif Intell ISSN: 2252-8938  1131

outputs a region proposal (set of locations of rectangles that can contain objects), along with the
corresponding rectangle's probability of containing the object.
RPN has 2 main steps:
− Feed forward images via DNN to get convolutional features (We can use the available backbone
networks such as VGG-16, ResNet50).
− Using the convolutional features slide up window: to create region proposals, we use a convolutional
layer with sliding windows called anchors. The output of this layer is the input of two fully-connected
layers that predict the location of the regions (Box-regression layer), as well as the probability of
containing the object (Box-classification) of that box as shown in Figure 2.

Figure 2. RPN architecture with anchor boxes [19]

2.1.2. Loss function

For each module in Faster R-CNN, we define a different loss function. With the Fast R-CNN
module, the loss used is traditional cross-entropy. Whereas for the RPN module, the loss function is defined
as follows,

1 𝜆
𝐿({𝑝𝑖 }, {𝑡𝑖 }) = ∑𝑖 𝐿𝑐𝑙𝑠 (𝑝𝑖 , 𝑝𝑖∗ ) + ∑𝑖 𝑝𝑖∗ 𝐿𝑟𝑒𝑔 (𝑡𝑖 , 𝑡𝐼∗ ) (1)
𝑁𝑐𝑙𝑠 𝑁𝑟𝑒𝑔

where 𝑖 is the index of the anchor in the mini-batch and 𝑝𝑖 is the predicted probability of anchor 𝑖
being an object. The ground-truth label value 𝑝𝑖∗ is 1 if the anchor is positive, and 0 when the anchor is
negative. 𝑡𝑖 is a 4-dimensional vector representing the predicted bounding box coordinates. 𝑡𝑖∗ is a
4-dimensional vector representing the coordinate value of the ground-truth box corresponding to the positive
anchor. 𝑁 is the number of anchor boxes taken out for consideration, 𝜆 is the equilibrium coefficient (=1).
𝐿𝑐𝑙𝑠 is the loss cross-entropy of 2 classes (Object and non-object), 𝐿𝑟𝑒𝑔 uses SmoothL1Loss.

2.2. PDIWS dataset

PDIWS [2] is a rare thermal image dataset for human detection in intrusion warning systems. The
special feature of this dataset is that it is artificial data, synthesized by the method of image editing in the
gradient domain. Each image in the dataset is made up of a human subject with an intrusive pose and a
background. The detailed specification of the dataset is described in Table 1. Accordingly, the data set
includes 2,000 training images and 500 test images in a ratio of 80:20. Note that the subjects in these two
subsets are completely distinct, so it ensures randomness when testing. In addition, the subjects are also
scaled in different proportions before being glued to the background, so it describes the various distances
from the camera to the subject. Figure 3(a) shows an example of a creeping subject, Figure 3(b) shows an
example of crawling subject, Figure 3(c) shows an example of stooping subject, Figure 3(d) shows an
example of a climbing subject, and Figure 3(e) show an example of other subject in the dataset. In this paper,
we divide the training set of the PDIWS dataset into two parts: training (1,750 images) and validation (250
images) at a ratio of 7:1, evenly distributed across classes. The purpose of this is to observe and evaluate the
training process to decide to stop at the right time to avoid overfitting.

Backbone search for object detection for applications in intrusion warning systems (Nguyen Duc Thuan)
1132  ISSN: 2252-8938

Table 1. PDIWS dataset description

Dataset specification
Total number of training images 2,000
Total number of test images 500
Number of classes 5
Number of subjects 1,000
Number of backgrounds 50
Image resolution 320x240
Synthesis method Poisson editing
Camera type Testo 875-2i
Field of view 32°x23°

Figure 3. Example subject images in PDIWS dataset. (a) Creeping, (b) Crawling, (c) Stooping,
(d) Climbing and (e) Other

3. METHOD
In this section, we describe in detail the architecture of the proposed model, the backbone width
search algorithm, and the criteria for method/model evaluation. An overview of the proposed method is
shown in Figure 4. The core part of the object detection algorithm is Faster R-CNN with the feature extractor
based on ResNet50 highlighted in orange. The neural network architecture search algorithm named random
width search is represented by curved orange arrows.

Figure 4. The architecture of the proposed backbone width search for object detection is based on Faster R-
CNN and ResNet50

3.1. Object detection architecture

The proposed object detection architecture is based on Faster R-CNN. We divide this architecture
into three parts for the convenience of description: feature extractor, region proposal network and output
classifier (including region of interest (ROI) pooling and classifier). First, the input image is passed through a
feature extractor to discover the features of the image. In the proposed architecture, the feature extractor is
chosen as ResNet50 because it is popular in the object detection task along with visual geometry group

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1129-1138

Int J Artif Intell ISSN: 2252-8938  1133

network (VGG). We add a factor called width scale (w) to modify ResNet50 architecture as shown in Table 2
with the note that the number of output channels is the smallest upper bound divisible by 8. The output of the
feature extractor is 2,048 feature maps with the size of each map reduced by 32 times in each dimension
compared to the input. This 2048 feature map will go in two directions, one goes into the region proposal
network and the other goes into the ROI pooling layer.

Table 2. Proposed ResNet50 architecture with width scale

Layer name Kernel size Output channels Number of blocks
conv1 7×7 64×w 1
1×1 64×w
conv2_x 3×3 64×w 3
1×1 256×w
1×1 128×w
conv3_x 3×3 128×w 4
1×1 512×w
1×1 256×w
conv4_x 3×3 256×w 6
1×1 1,024×w
1×1 512×w
conv5_x 3×3 512×w 3
1×1 2,048×w
fully-connected - 1,000×w 1
fully-connected - 5 1

At the region proposal network, at each pixel in the feature map, it is further fed into the convolution
layer to generate feature vectors of length 256. The number of feature vectors is equal to the number of
predefined anchor boxes. Then, a fully-connected layer is used to predict the object's presence and bounding
box according to each anchor box. These predictions are then fed into ROI pooling to remove overlapping
regions of the same object. An important point in the RPN network is the anchor box. They are usually
preselected according to the size and proportions of the objects contained in the training data. In the proposed
method, the recommended anchor boxes are 15 types with aspect ratios of 0.5, 1.0, and 2.0.
Finally, the output classifier with ROI pooling and classifier is activated. The predicted bounding
boxes then capture corresponding regions in the 2,048 feature maps. The non-maximum suppression (NMS)
algorithm is then utilized to remove boxes containing the same object. Then the filtered boxes are reshaped to
7×7 by the ROI pooling layer before passing through the classifier. The classifier is a fully-connected layer
that projects the class of the object. Loss functions for the proposed method are the same as the original faster
R-CNN described in section 2.1.

3.2. Backbone width search algorithm

This is the most important part of the proposed method. As argued in the Introduction section,
scaling the neural network in breadth is advantageous for inserting feature pyramid networks, which have
recently been widely advertised. Therefore, we propose a simple solution to scale the width with the essence
of changing the number of output channels of each hidden layer in the neural network. Our algorithm is
described in Algorithm 1 as follows,

Algorithm 1: Backbone width search

input: width_scale ~ U{num_of_trial, 0;
1}, num_of_epoch,
model=Faster_RCNN, backbone_hist=[ ]
for interation = 1 to num_of_trial do:
backbone.layer.out_channel *=
width_scale[interation]
if backbone in backbone_hist then:
continue
end if
backbone_hist.add(backbone)
for epoch = 1 to num_of_epoch do:
loss_dict = model.train()
mAP, num_of_param, MACs = model.eval()
model.save()
log.save(loss_dict, mAP, num_of_param,
MACs)
end for
end for
output: model, log
Backbone search for object detection for applications in intrusion warning systems (Nguyen Duc Thuan)
1134  ISSN: 2252-8938

Accordingly, we have to predetermine the number of trials (num_of_trial) and select width scales
following the uniform distribution U(0;1). At each value of the width scale, we rebuild the backbone by
multiplying the number of output channels of each hidden layer with the width scale. The resulting network
architecture is saved to check for duplicates in subsequent attempts. Then the model is trained from scratch
with the new architecture and then the model along with the loss dictionary, mAP, number of parameters, and
number of MACs values will be saved. In this paper, we choose 10 trials for simplification. The
configuration parameters for the training process will be presented in the results section.

3.3. Evaluation criteria

3.3.1. Mean average precision (mAP)
In binary classification, the model assigns prediction scores to samples. Categorization into classes
depends on a set threshold. If the score meets or exceeds the threshold, the sample is in one class; otherwise,
it's in the other. Positive classification occurs when the score is at or above the threshold; negative
classification applies when the score falls below it. We use the terms precision and recall to evaluate the
performance of the binary classification for each class as shown in Figure 5,
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒
𝑅𝑒𝑐𝑎𝑙𝑙 = (2)
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒+𝐹𝑎𝑙𝑠𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒

𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (3)
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒+𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒

Figure 5. Illustration of binary classification predictions

Of course, we want a model to have high precision and high recall. To take into account both
metrics, we use the precision-recall curve, which is a plot of the precision (y-axis) and the recall (x-axis) for
different probability thresholds. Then we compute average precision (AP) by averaging the precision values
on the precision-recall curve where the recall is in the range [0,1],

𝑖
𝐴𝑃 = ∑𝑁
𝑖=0 𝑝 ( ) (4)
𝑁

where N denotes the number of thresholds and p(r) denotes the value of the precision when the
recall is equal to r. The mAP is calculated by the mean of AP values over all classes.

3.3.2. Number of parameters and multiply-accumulate operations (MACs)

For a fully-connected layer, given the number of input neurons is a and the number of output is b. In
this case, the number of parameters and MACs is determined by,

𝑃𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠𝐹𝐶 = 𝑎 × 𝑏 (5)

𝑀𝐴𝐶𝑠𝐹𝐶 = 𝑎 × 𝑎 × 𝑏 (6)

For a 2D convolutional layer, given the input is a×a×n, output of b×b×m and kernel size is k×k.
Then the two terms parameters and MACs are defined by,

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1129-1138

Int J Artif Intell ISSN: 2252-8938  1135

𝑃𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠𝐶𝑁𝑁 = k × k × n × m (7)

𝑀𝐴𝐶𝑠𝐶𝑁𝑁 = k × k × n × b × b × m (8)

4. RESULTS AND DISCUSSION

Our experiments are performed on the PDIWS dataset with the following training parameters:
number of epochs (100), batch size (2), learning rate (0.005), momentum (0.9), optimizer (Adam). They are
executed using Pytorch framework on a computer with Intel i7 12700F CPU, 16 GB RAM, and 24GB Nvidia
GeForce GTX 3090 GPU. We first present the results of the backbone architecture search Table 3 with the
criteria mAP_50 is mAP at IoU=0.5, mAP_75 is mAP at IoU=0.75 and mAP_coco is the average mAP with
IoU=[0.5, 0.95] with step 0.05. As expected, architectures with small-width scales perform worse than larger
ones. This can be explained by the small size of the model, the number of neurons is not enough to learn the
necessary features. Meanwhile, network architectures larger than a certain threshold will not improve
performance. Specifically, width scale values greater than 0.5 results in mAP_50 not being too different. The
mAP value increases gradually as the width scale increases but does not exceed 47.55%, the increased
amplitude is only approximately 2%. This trend is also present in mAP_75 and mAP_coco. Meanwhile, with
a width scale of less than 0.5, the survey mAP values decrease exponentially as the width scale decreases
from 0.5 to 0.1. On a scale of 0.1, the value of mAP_50 is only 18.17%, mAP_75 is 16.05%, mAP_coco is
13.54%, all extremely bad. In addition, for the number of parameters and MACs, the architectures searched
had the number of parameters in million (M) exponential to the width scale (0.55M at 0.1 and 55M at 0.1).
The number of MAC operations follows a similar trend, and varies in the form of an exponential function.

Table 3. Performance comparison between different searched model width

Model width 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
mAP_50 (%) 18.17 30.24 39.29 42.71 45.23 45.42 45.78 46.07 46.28 47.55
mAP_75 (%) 16.05 28.4 37.85 41.01 43.32 43.61 44.00 43.98 44.78 46.24
mAP_coco (%) 13.54 24.87 34.76 38.33 40.51 40.36 40.97 40.86 41.85 42.61
Parameters (M) 0.55 2.12 4.99 8.89 13.79 19.76 26.99 35.35 44.82 55.15
MACs (B) 0.08 0.28 0.62 1.13 1.89 2.89 4.09 5.54 7.31 9.48

Visualization of the table results can be found in Figure 6. Here, we can easily observe the trend of
the criteria when the width scale changes. In order to choose a suitable scale, some threshold values must be
predefined. In this paper, we think that the mAP_50 value must be above 45% for the model to be called
good because this is the near-convergence value of the model. Therefore, from Figure 6, we choose a width
scale value of 0.5 as the best value with mAP_50 reaching 45.42%, mAP_75 reaching 43.32%, mAP_coco
reaching 40.51%, the number of parameters is 13.79M and the number of MAC operations is 1.89 billion (B).
In addition, we can also choose another threshold value such as model size if the hardware is constrained or
inference time if it requires real-time calculation.

Figure 6. Visualization of search results

Backbone search for object detection for applications in intrusion warning systems (Nguyen Duc Thuan)
1136  ISSN: 2252-8938

To increase the credibility of the obtained model as well as of the proposed method, comparative
experiments were conducted. Table 4 summarizes the mAP values (at IoU=0.5), the number of parameters,
MACs and inference time of various methods/models with different backbones. Specifically, the
methods/models selected for the survey are well-known names such as single shot detection (SSD) [24], fully
convolutional one-stage (FCOS) [25], you only live once version 3 (YOLOv3) [26]. Following them are
popular backbones such as VGG-16 [27], ResNet50 [20], and MobileNet-V2 [28]. The results show that the
criteria have trade-offs, so we should have a metric to compare the methods and choose the best one. Figure 7
shows some example results on test set with labeled bounding boxes over all classes.

Table 4. Comparison between the obtained model with other object detection methods
Method Backbone mAP (%) Parameters (M) MACs (B) Inference (s)
Faster R-CNN VGG-16 45.76 23.92 41.45 0.647
Faster R-CNN MobileNet-V2 37.48 4.35 1.64 0.036
Faster R-CNN ResNet50 47.55 55.15 9.48 0.216
SSD VGG-16 40.48 23.84 41.12 0.374
SSD MobileNet-V2 34.73 4.21 1.45 0.027
SSD ResNet50 43.39 54.84 9.22 0.127
FCOS VGG-16 41.73 23.81 41.06 0.329
FCOS MobileNet-V2 33.94 4.18 1.42 0.018
FCOS ResNet50 43.43 54.85 9.13 0.105
YOLOv3 VGG-16 44.17 23.72 40.85 0.246
YOLOv3 MobileNet-V2 36.76 4.12 1.32 0.012
YOLOv3 ResNet50 47.15 54.62 8.94 0.062
0.5 ResNet50 45.23 13.79 1.89 0.068
Proposed
0.6 ResNet50 45.42 19.76 2.89 0.071

Figure 7. Examples of predicted bounding boxes over five classes in a test set

5. CONCLUSION
We have proposed a new and simple backbone architecture search method for object detection based
on Faster R-CNN and ResNet50. The proposed method is performed on the PDIWS dataset so that it can be
deployed on intrusion warning systems. The obtained model exhibits object detection with over 45% mAP
with only 13.79M parameters and 1.89B MAC operations, outperforming other model scales. Comparative
experiments with classical methods are also performed to evaluate the value of the obtained model. In
summary, this work has contributed to supplementing theory and experimental results for the trend of object
detection model optimization.

ACKNOWLEDGEMENTS
Nguyen Duc Thuan was funded by the Master, PhD Scholarship Programme of Vingroup
Innovation Foundation (VINIF), code VINIF.2022.ThS.086.

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1129-1138

Int J Artif Intell ISSN: 2252-8938  1137

REFERENCES
[1] N. D. Thuan, “Backbone-width-search-Faster-RCNN,” [Online]. Available: https://github.com/thuan-researcher/Backbone-width-
search-Faster-RCNN.
[2] N. D. Thuan, L. H. Anh, and H. S. Hong, “PDIWS: Thermal imaging dataset for person detection in intrusion warning systems,”
IEEE Workshop on Statistical Signal Processing Proceedings, vol. 2023-July, pp. 71–75, 2023, doi:
10.1109/SSP53291.2023.10208055.
[3] N. U. Huda, B. D. Hansen, R. Gade, and T. B. Moeslund, “The effect of a diverse dataset for transfer learning in thermal person
detection,” Sensors (Switzerland), vol. 20, no. 7, 2020, doi: 10.3390/s20071982.
[4] G. Wang et al., “BED: A real-time object detection system for edge devices,” International Conference on Information and
Knowledge Management, Proceedings, pp. 4994–4998, 2022, doi: 10.1145/3511808.3557168.
[5] N. D. Thuan, T. P. Dong, B. Q. Manh, H. A. Thai, T. Q. Trung, and H. S. Hong, “Edge-focus thermal image super-resolution
using generative adversarial network,” 2022 International Conference on Multimedia Analysis and Pattern Recognition, MAPR
2022 - Proceedings, 2022, doi: 10.1109/MAPR56351.2022.9924742.
[6] X. He, K. Zhao, and X. Chu, “AutoML: A survey of the state-of-the-art,” Knowledge-Based Systems, vol. 212, 2021, doi:
10.1016/j.knosys.2020.106622.
[7] P. Ren et al., “A comprehensive survey of neural architecture search: Challenges and solutions,” ACM Computing Surveys, vol.
54, no. 4, 2021, doi: 10.1145/3447582.
[8] Y. Chen, T. Yang, X. Zhang, G. Meng, X. Xiao, and J. Sun, “DetNAS: Backbone search for object detection,” Advances in
Neural Information Processing Systems, vol. 32, 2019.
[9] X. Du et al., “SpineNet: Learning scale-permuted backbone for recognition and localization,” Proceedings of the IEEE Computer
Society Conference on Computer Vision and Pattern Recognition, pp. 11589–11598, 2020, doi:
10.1109/CVPR42600.2020.01161.
[10] G. Ghiasi, T. Y. Lin, and Q. V. Le, “NAS-FPN: Learning scalable feature pyramid architecture for object detection,” Proceedings
of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2019-June, pp. 7029–7038, 2019,
doi: 10.1109/CVPR.2019.00720.
[11] H. Xu, L. Yao, Z. Li, X. Liang, and W. Zhang, “Auto-FPN: Automatic network architecture adaptation for object detection
beyond classification,” Proceedings of the IEEE International Conference on Computer Vision, vol. 2019-October, pp. 6648–
6657, 2019, doi: 10.1109/ICCV.2019.00675.
[12] L. Yao, H. Xu, W. Zhang, X. Liang, and Z. Li, “SM-NAS: Structural-to-modular neural architecture search for object detection,”
AAAI 2020 - 34th AAAI Conference on Artificial Intelligence, pp. 12661–12668, 2020, doi: 10.1609/aaai.v34i07.6958.
[13] N. D. Thuan, N. T. Hue, P. Q. Vuong, and H. S. Hong, “Intelligent bearing fault diagnosis with a lightweight neural network,”
2022 11th International Conference on Control, Automation and Information Sciences, ICCAIS 2022, pp. 261–266, 2022, doi:
10.1109/ICCAIS56082.2022.9990211.
[14] Y. He, P. Liu, Z. Wang, Z. Hu, and Y. Yang, “Filter pruning via geometric median for deep convolutional neural networks
acceleration,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2019-
June, pp. 4335–4344, 2019, doi: 10.1109/CVPR.2019.00447.
[15] H. Li, H. Samet, A. Kadav, I. Durdanovic, and H. P. Graf, “Pruning filters for efficient convnets,” 5th International Conference
on Learning Representations, ICLR 2017 - Conference Track Proceedings, 2017.
[16] Z. Liu, J. Li, Z. Shen, G. Huang, S. Yan, and C. Zhang, “Learning efficient convolutional networks through network slimming,”
Proceedings of the IEEE International Conference on Computer Vision, vol. 2017-Octob, pp. 2755–2763, 2017, doi:
10.1109/ICCV.2017.298.
[17] B. Jacob et al., “Quantization and training of neural networks for efficient integer-arithmetic-only inference,” Proceedings of the
IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 2704–2713, 2018, doi:
10.1109/CVPR.2018.00286.
[18] M. Courbariaux, I. Hubara, D. Soudry, R. El-Yaniv, and Y. Bengio, “Binarized neural networks: training deep neural networks
with weights and activations constrained to +1 or -1,” 2016, [Online]. Available: http://arxiv.org/abs/1602.02830.
[19] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal networks,” IEEE
Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137–1149, 2017, doi:
10.1109/TPAMI.2016.2577031.
[20] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE Computer Society
Conference on Computer Vision and Pattern Recognition, vol. 2016-December, pp. 770–778, 2016, doi: 10.1109/CVPR.2016.90.
[21] M. Tan and Q. V. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” 36th International Conference
on Machine Learning, ICML 2019, vol. 2019-June, pp. 10691–10700, 2019.
[22] T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature Pyramid Networks for Object Detection,”
Proceedings - 2019 IEEE Intl Conf on Parallel and Distributed Processing with Applications, Big Data and Cloud Computing,
Sustainable Computing and Communications, Social Computing and Networking, ISPA/BDCloud/SustainCom/SocialCom 2019,
pp. 1500–1504, Dec. 2016, doi: 10.1109/ISPA-BDCloud-SustainCom-SocialCom48970.2019.00217.
[23] X. Xie, Y. Ma, B. Liu, J. He, S. Li, and H. Wang, “A deep-learning-based real-time detector for grape leaf diseases using
improved convolutional neural networks,” Frontiers in Plant Science, vol. 11, 2020, doi: 10.3389/fpls.2020.00751.
[24] W. Liu et al., “SSD: Single shot multibox detector,” Lecture Notes in Computer Science (including subseries Lecture Notes in
Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9905 LNCS, pp. 21–37, 2016, doi: 10.1007/978-3-319-46448-
0_2.
[25] Z. Tian, C. Shen, H. Chen, and T. He, “FCOS: Fully convolutional one-stage object detection,” Proceedings of the IEEE
International Conference on Computer Vision, vol. 2019-October, pp. 9626–9635, 2019, doi: 10.1109/ICCV.2019.00972.
[26] J. Redmon and A. Farhadi, “YOLOv3: An incremental improvement,” 2018, [Online]. Available: http://arxiv.org/abs/1804.02767.
[27] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” 3rd International
Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings, 2015.
[28] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. C. Chen, “MobileNetV2: Inverted residuals and linear bottlenecks,”
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 4510–4520, 2018, doi:
10.1109/CVPR.2018.00474.

Backbone search for object detection for applications in intrusion warning systems (Nguyen Duc Thuan)
1138  ISSN: 2252-8938

BIOGRAPHIES OF AUTHORS

Nguyen Duc Thuan received his B.S. degree from Hanoi University of Science
and Technology (Vietnam) in 2021. He is pursuing an M.S. degree in Electrical and
Electronic Engineering at Hanoi University of Science and Technology. His current major
research fields include Signal Processing, Energy Harvesting, Fault Detection and Embedded
Artificial Intelligence (EAI). He can be contacted at email: [email protected].

Nguyen Thi Lan Huong received her M.S. degree in Measurement and
Information from Kharkov Polytechnic University (Ukraine) in 1996 and her Ph.D. in
Electrical Measurement from Hanoi University of Science and Technology (Vietnam) in
2005. From 1996 until now, she has been a lecturer at the School of Electrical and Electronic
Engineering, Hanoi University of Science and Technology. The main research direction is the
measurement and control systems, measurement signal processing, and research on sensory
environments. She can be contacted at email: [email protected].

Hoang Si Hong received his B.E. and M.E. degrees in Electrical Engineering
from Hanoi University of Technology, Hanoi, Vietnam, in 1999 and 2001, respectively, and
his Ph.D. degree from the University of Ulsan, Ulsan, South Korea, in 2010. Now, he is an
Associate Professor at the School of Electrical and Electronic Engineering, Hanoi University
of Science and Technology. His research interests include piezoelectric material, gas sensors,
Graphene-based on SAW sensors, wireless sensors and harvesting energy and advanced
measurement using Artificial Intelligence technology. He can be contacted at email:
[email protected].

Int J Artif Intell, Vol. 13, No. 1, March 2024: 1129-1138