<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>Aiming at the problems of low object detection accuracy due to complex background and insufficient semantic information of shallow features in the object detection SSD algorithm, this paper improves the existing SSD algorithm. First, the original vgg16 network is replaced by the ResNet50 network, and the residual network structure as well as the Batch Normalization layer are added, which are used to improve the accuracy of the feature extraction network; Second, a feature fusion module is designed to fuse adjacent feature maps to improve the detection effect by integrating contextual information; Third, the SE attention mechanism is introduced to give channel weights adaptively and enhance the useful feature channels; Finally, the object detection analysis experiments are conducted on the PASCAL VOC2012 dataset. The experimental results show that the improved SSD algorithm in this paper is able to achieve an mean average precision of 72.7% in the data set, which is 2.1% better than the original SSD-VGG16 and greatly improves the object detection effect.</p>
</abstract>

Aiming at the problems of low object detection accuracy due to complex background and insufficient semantic information of shallow features in the object detection SSD algorithm, this paper improves the existing SSD algorithm. First, the original vgg16 network is replaced by the ResNet50 network, and the residual network structure as well as the Batch Normalization layer are added, which are used to improve the accuracy of the feature extraction network; Second, a feature fusion module is designed to fuse adjacent feature maps to improve the detection effect by integrating contextual information; Third, the SE attention mechanism is introduced to give channel weights adaptively and enhance the useful feature channels; Finally, the object detection analysis experiments are conducted on the PASCAL VOC2012 dataset. The experimental results show that the improved SSD algorithm in this paper is able to achieve an mean average precision of 72.7% in the data set, which is 2.1% better than the original SSD-VGG16 and greatly improves the object detection effect.

SSD Object Detection Algorithm Based on Feature Fusion and Channel Attention

School of Computer Science and Engineering

International Journal of Advanced Network, Monitoring and Controls

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

{"article-title":"SSD Object Detection Algorithm Based on Feature Fusion and Channel Attention"}

Aiming at the problems of low object detection accuracy due to complex background and insufficient semantic information of shallow features in the object...

Hardware	Processor Video Cards	Intel(R)Core(TM) i7-6500U GeForce_RTX_2080_Ti
Software	Operating System	windows10
	Deep Learning Framework	pytorch-gpu
	Compiler Language	python
	Compilers	pycharm

Feature Map	Width and height of the feature map	Default boxes size	Number of default boxes
Feature Map1	38 × 38	21{1/2,1,2}; $\sqrt{21 \times 45}$ \sqrt {21\, \times 45} {1}	38 × 38 × 4
Feature Map2	19 × 19	45{1/3,1/2,1,2,3}; $\sqrt{45 \times 99}$ \sqrt {45\, \times 99} {1}	19 × 19 × 6
Feature Map3	10 × 10	99{1/3,1/2,1,2,3}; $\sqrt{99 \times 153}$ \sqrt {99\, \times 153} {1}	10 × 10 × 6
Feature Map4	5 × 5	153{1/3,1/2,1,2,3}; $\sqrt{153 \times 207}$ \sqrt {153\, \times 207} {1}	5 × 5 × 6
Feature Map5	3 × 3	207{1/2,1,2}; $\sqrt{207 \times 261}$ \sqrt {207\, \times 261} {1}	3 × 3 × 4
Feature Map6	1 × 1	261{1/2,1,2}; $\sqrt{261 \times 315}$ \sqrt {261\, \times 315} {1}	1 × 1 × 4

Algorithm	mAP
SSD-VGG16	70.6%
SSD-ResNet50	72.1%
The improved algorithm	72.7%

Parameter	Value
learning rate	0.0005
momentum	0.9
weight_decay	0.0005
batch size	16
epoch	50
step_size	5

SSD Object Detection Algorithm Based on Feature Fusion and Channel Attention

Published Online: May 24, 2023

Page range: 80 - 89

DOI: https://doi.org/10.2478/ijanmc-2022-0029

Keywords
Object Detection, Feature Fusion, Attention Mechanism, SSD Algorithm

© 2022 Leilei Fan et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Environment configuration table

Default dimension and quantity of feature map

Comparison of detection results of different algorithms on the PASCAL VOC2012 dataset

Training parameters setting table

SSD Object Detection Algorithm Based on Feature Fusion and Channel Attention

Published Online: May 24, 2023

Page range: 80 - 89

DOI: https://doi.org/10.2478/ijanmc-2022-0029

KeywordsObject Detection, Feature Fusion, Attention Mechanism, SSD Algorithm

© 2022 Leilei Fan et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Environment configuration table

Default dimension and quantity of feature map

Comparison of detection results of different algorithms on the PASCAL VOC2012 dataset

Training parameters setting table

Keywords
Object Detection, Feature Fusion, Attention Mechanism, SSD Algorithm