Research on Lightweight Method of Insulator Target Detection Based on Improved SSD
Abstract
:1. Introduction
2. SSD Network Model
3. Improved SSD Network Model
- (1)
- The original feature extraction network, VGG-16, is replaced with a lightweight Ghost Module network to initially achieve model lightweighting.
- (2)
- The Neck part of the SSD network adopts an FPN+PAN structure to enhance feature extraction capabilities. To facilitate the fusion of local and global features, a SimSPPF structure is introduced at each input end of the Neck.
- (3)
- Multiple Spatial and Channel Squeeze-and-Excitation (scSE) attention mechanism modules are incorporated into the Neck section, enabling the network to better focus on channels containing critical feature information while preserving positional information of feature layers.
- (4)
- The original six detection heads are reduced to four to accelerate the network’s inference speed. To improve the recognition of occluded and overlapping objects, DIoU-NMS replaces the conventional non-maximum suppression.
- (5)
- Channel pruning strategies are employed to eliminate unimportant weight matrices, further lightweighting the constructed network model and achieving model compression objectives.
- (6)
- To mitigate the impact of channel pruning on detection accuracy, knowledge distillation is applied to fine-tune the lightweight network model, ensuring detection precision is maintained.
3.1. Feature Extraction Network
3.2. Feature Fusion Network
3.3. Model Compression and Fine-Tuning
4. Experimental Results and Analysis
4.1. Experimental Environment
4.2. Datasets and Training
Algorithm 1: pseudo-code of the proposed model |
Input: An image to be detected |
Output: An image with detection results |
1: Resize the input image to 416 × 416 and normalize it. |
2: Pass the processed image through the backbone network to extract features. |
3: Feed the extracted features into the network model (backbone, neck, and head) to obtain candidate bounding boxes. |
4: For each candidate bounding box: Perform classification and bounding box regression; Decode the regression results to determine the final position of the bounding box; Apply DIoU-NMS to filter out overlapping detections; Map the detection result onto the original image. |
5: Return the image with the overlaid detection results. |
4.3. Evaluating Indicator
4.4. Ablation Experiment
4.5. Comparison of Different Algorithm Effects
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Wei, L.; Jin, J.; Deng, K.; Liu, H. Insulator defect detection in transmission line based on an improved lightweight YOLOv5s algorithm. Electr. Power Syst. Res. 2024, 233, 110464. [Google Scholar] [CrossRef]
- Qi, Y.; Mu, S.; Wang, J.; Wang, L. Intelligent Recognition of Transmission Line Inspection Image Based on Deep Learning. J. Phys. Conf. Ser. 2021, 1757, 012056. [Google Scholar] [CrossRef]
- Shahrzad, F.; Azam, K.; Hossein, N. PTSRGAN: Power transmission lines single image super-resolution using a generative adversarial network. Int. J. Electr. Power Energy Syst. 2024, 155, 109607. [Google Scholar]
- Girshick, R.; Donahue, J.; Darrell, T.; Malik, J. Region-Based Convolutional Networks for Accurate Object Detection and Segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 2015, 38, 142–158. [Google Scholar] [CrossRef]
- Arora, N.; Kumar, Y.; Karkra, R.; Kumar, M. Automatic vehicle detection system in different environment conditions using fast R-CNN. Multimed. Tools Appl. 2022, 81, 18715–18735. [Google Scholar] [CrossRef]
- Zhao, W.; Xu, M.; Cheng, X.; Zhao, Z. An insulator in transmission lines recognition and fault detection model based on improved faster RCNN. IEEE Trans. Instrum. Meas. 2021, 70, 5016408. [Google Scholar] [CrossRef]
- Zhao, Z.; Zhen, Z.; Zhang, L.; Qi, Y.; Kong, Y.; Zhang, K. Insulator Detection Method in Inspection Image Based on Improved Faster R-CNN. Energies 2019, 12, 1204. [Google Scholar] [CrossRef]
- Haijian, H.; Yicen, L.; Haina, R. Detection of Insulators on Power Transmission Line Based on an Improved Faster Region-Convolutional Neural Network. Algorithms 2022, 15, 83. [Google Scholar] [CrossRef]
- Redmon, J.; Divvala, K.S.; Girshick, B.R.; Farhadi, A. You Only Look Once: Unified, Real-Time Object Detection. arXiv 2015, arXiv:1506.02640. [Google Scholar]
- Gu, J.; Hu, J.; Jiang, L.; Wang, Z.; Zhang, X.; Xu, Y.; Zhu, J.; Fang, L. Research on Object Detection of Overhead Transmission Lines Based on Optimized YOLOv5s. Energies 2023, 16, 2706. [Google Scholar] [CrossRef]
- Wang, T.; Zhai, Y.; Li, Y.; Wang, W.; Ye, G.; Jin, S. Insulator Defect Detection Based on ML-YOLOv5 Algorithm. Sensors 2023, 24, 204. [Google Scholar] [CrossRef]
- Liu, W.; Anguelov, D.; Erhan, D.; Szegedy, C.; Reed, S.; Fu, C.-Y.; Berg, A.C. SSD: Single Shot MultiBox Detector. arXiv 2015, arXiv:1512.02325. [Google Scholar]
- Liu, X.; Li, Y.; Shuang, F.; Gao, F.; Zhou, X.; Chen, X. ISSD: Improved SSD for Insulator and Spacer Online Detection Based on UAV System. Sensors 2020, 20, 6961. [Google Scholar] [CrossRef] [PubMed]
- Liu, J.; Hu, M.; Dong, J.; Lu, X. The application of a lightweight model FA-YOLOv5 with fused attention mechanism in insulator defect detection. Front. Energy Res. 2023, 11, 1283394. [Google Scholar] [CrossRef]
- Deng, X.; Li, S. An Improved SSD Object Detection Algorithm Based on Attention Mechanism and Feature Fusion. J. Phys. Conf. Ser. 2023, 2450, 012088. [Google Scholar] [CrossRef]
- Zhai, S.; Shang, D.; Wang, S.; Dong, S. DF-SSD: An Improved SSD Object Detection Algorithm Based on DenseNet and Feature Fusion. IEEE Access 2020, 8, 24344–24357. [Google Scholar] [CrossRef]
- Qian, H.; Wang, H.; Feng, S.; Yan, S. FESSD:SSD target detection based on feature fusion and feature enhancement. J. Real-Time Image Process. 2023, 20, 2. [Google Scholar] [CrossRef]
- Hong, F.; Lu, C.H.; Tao, W.; Jiang, W. Improved SSD Model for Pedestrian Detection in Natural Scene. Wirel. Commun. Mob. Comput. 2022, 2022, 1500428. [Google Scholar] [CrossRef]
- Han, K.; Wang, Y.; Tian, Q.; Guo, J.; Xu, C.; Xu, C. GhostNet: More features from cheap operations. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, 13–19 June 2020; pp. 1577–1586. [Google Scholar]
- Li, L.; Zheng, C.; Mao, C.; Deng, H.; Jin, T. Scale-Insensitive Object Detection via Attention Feature Pyramid Transformer Network. Neural Process. Lett. 2021, 54, 581–595. [Google Scholar] [CrossRef]
- Hu, H.; Zhu, Z. Sim-YOLOv5s: A method for detecting defects on the end face of lithium battery steel shells. Adv. Eng. Inform. 2023, 55, 101824. [Google Scholar] [CrossRef]
- Nie, P.; Guo, Y.; Lou, B.; Yang, C.; Cao, L.; Pan, W. Tool wear monitoring based on scSE-ResNet-50-TSCNN model integrating machine vision and force signals. Meas. Sci. Technol. 2024, 35, 086117. [Google Scholar] [CrossRef]
- Yan, P.; Sun, Q.; Yin, N.; Hua, L.; Shang, S.; Zhang, C. Detection of coal and gangue based on improved YOLOv5.1 which embedded scSE module. Measurement 2022, 188, 110530. [Google Scholar]
- Shepley, A.; Falzon, G.; Kwan, P. Confluence: A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection. IEEE Trans. Pattern Anal. Mach. Intell. 2023, 10, 11561–11574. [Google Scholar] [CrossRef]
- Jiang, L.; Nie, W.; Zhu, J.; Gao, X.; Lei, B. Lightweight object detection network model suitable for indoor mobile robots. J. Mech. Sci. Technol. 2022, 36, 907–920. [Google Scholar] [CrossRef]
- He, Y.; Dong, X.; Kang, G.; Fu, Y.; Yan, C.; Yang, Y. Asymptotic Soft Filter Pruning for Deep Convolutional Neural Networks. IEEE Trans. Cybern. 2020, 50, 3594–3604. [Google Scholar] [CrossRef]
- Huang, M.; Liu, Y.; Zhao, L.; Wang, G. A lightweight deep neural network model and its applications based on channel pruning and group vector quantization. Neural Comput. Appl. 2023, 36, 5333–5346. [Google Scholar] [CrossRef]
- Zhang, Z.; Kong, S.; Peng, K. Improved YOLOv4 Power Insulator Fault Detection. J. Phys. Conf. Ser. 2021, 2010, 012148. [Google Scholar] [CrossRef]
- Arima, K.; Nagata, F.; Shimizu, T.; Otsuka, A.; Kato, H.; Watanabe, K.; Habib, M.K. Improvements of detection accuracy and its confidence of defective areas by YOLOv2 using a data set augmentation method. Artif. Life Robot. 2023, 28, 625–631. [Google Scholar]
- Zhao, Z.; Lv, X.; Xi, Y.; Miao, S. Defect detection method for key area guided transmission line components based on knowledge distillation. Front. Energy Res. 2023, 11, 1287024. [Google Scholar]
- Zhao, Z.; Lyu, J.; Chu, Y.; Liu, K. Toward generalizable robot vision guidance in real-world operational manufacturing factories: A Semi-Supervised Knowledge Distillation approach. Robot. Comput. Integr. Manuf. 2024, 86, 102639. [Google Scholar] [CrossRef]
- Du, W.; Geng, L.; Zhao, Z.; Wang, C.; Huo, J. Decoupled knowledge distillation method based on meta-learning. High-Confid. Comput. 2024, 4, 100164. [Google Scholar] [CrossRef]
- Wei, L.; Tong, Y. Enhanced-YOLOv8: A new small target detection model. Digit. Signal Process. 2024, 153, 104611. [Google Scholar] [CrossRef]
- Li, Y.; Li, J.; Zhai, Y.; Meng, P. Detection of Self-explosive Insulators in Aerial Images Based on Improved YOLO v4. J. Phys. Conf. Ser. 2022, 2320, 012025. [Google Scholar]
Symbol | Description |
---|---|
the scale of candidate boxes for the m-th feature map | |
the maximum scale of candidate boxes | |
the minimum scale of candidate boxes | |
the total number of feature maps | |
h | the picture height |
w | the picture width |
c | the picture length |
channel feature map | |
undergoes a linear operation | |
, | the size of the linear operation kernel |
the compressed feature maps are then normalized by the sigmoid function | |
the ReLU function | |
M | the prediction box with a higher prediction score |
the other prediction boxes | |
the Euclidean distance between M and | |
non-zero constant | |
the input to the BN layer | |
Y | the output from the BN layer |
represents the normalized scale factor | |
the variance computed over a mini-batch for the BN layer | |
the mean computed over a mini-batch for the BN layer | |
a bias compensation in the normalization process | |
encompassing all prunable channels | |
FPS | Frame Per Second |
Tp | true positive predictions |
Fp | false positive predictions |
Fn | indicates false negative predictions |
Category | Parameter |
---|---|
CPU | 12th Gen Intel(R) Core(TM) i7-12700KF 3.6 GHz |
Memory | 32 G |
GPU | NVIDIA GeForce RTX 3090Ti |
GPU memory | 24 G |
OS | Windows 11 |
CUDA version | CUDA 11.0 |
cuDNN | cuDNN 7.6.5 |
Language | Python 3.6 |
Models | VGG-16 | Ghost Module | SimSPPF | scSE | FPN+PAN | Lightweight |
---|---|---|---|---|---|---|
SSD | √ | × | × | × | × | × |
A | × | √ | × | × | × | × |
B | × | √ | √ | × | × | × |
C | × | √ | √ | √ | × | × |
D | × | √ | √ | √ | √ | √ |
Models | Parameters | FLOPs | P | R | FPS f/s | [email protected]/% |
---|---|---|---|---|---|---|
SSD | 26.15 M | 118.95 G | 0.95 | 0.81 | 132 | 96.8% |
A | 5.07 M | 3.21 G | 0.96 | 0.79 | 111 | 93.2% |
B | 5.20 M | 3.38 G | 0.97 | 0.77 | 101 | 94.8% |
C | 7.04 M | 3.39 G | 0.94 | 0.81 | 101 | 95.6% |
D | 0.61 M | 1.49 G | 0.78 | 1 | 67 | 98.0% |
Models | Lightweight | FPS/f/s | P | R | Parameters | FLOPs | [email protected]/% |
---|---|---|---|---|---|---|---|
SSD | 132 | 0.95 | 0.81 | 26.15 M | 118.95 G | 96.8% | |
YOLOv3 | 71 | 0.80 | 0.98 | 61.52 M | 65.60 G | 96.9% | |
YOLOv5 | 94 | 0.94 | 0.99 | 47.06 M | 115.92 G | 98.2% | |
Faster-RCNN | 22 | 0.70 | 1 | 137.10 M | 370.21 G | 98.6% | |
Ghost-YOLOv3 | √ | 55 | 0.77 | 0.98 | 46.45 M | 25.32 G | 95.5% |
YOLOv3-Tiny | √ | 142 | 0.62 | 0.81 | 8.67 M | 5.49 G | 73.8% |
Ours | √ | 67 | 0.78 | 1 | 0.61 M | 1.49 G | 98.0% |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Zeng, B.; Zhou, Y.; He, D.; Zhou, Z.; Hao, S.; Yi, K.; Li, Z.; Zhang, W.; Xie, Y. Research on Lightweight Method of Insulator Target Detection Based on Improved SSD. Sensors 2024, 24, 5910. https://doi.org/10.3390/s24185910
Zeng B, Zhou Y, He D, Zhou Z, Hao S, Yi K, Li Z, Zhang W, Xie Y. Research on Lightweight Method of Insulator Target Detection Based on Improved SSD. Sensors. 2024; 24(18):5910. https://doi.org/10.3390/s24185910
Chicago/Turabian StyleZeng, Bing, Yu Zhou, Dilin He, Zhihao Zhou, Shitao Hao, Kexin Yi, Zhilong Li, Wenhua Zhang, and Yunmin Xie. 2024. "Research on Lightweight Method of Insulator Target Detection Based on Improved SSD" Sensors 24, no. 18: 5910. https://doi.org/10.3390/s24185910