-
摘要: 在輸電場景中，吊車等大型機械的運作會威脅到輸電線路的安全。針對此問題，從訓練數據、網絡結構和算法超參數的角度進行研究，設計了一種新的端到端的輸電線路威脅檢測網絡結構TATLNet，其中包括可疑區域生成網絡VRGNet和威脅判別網絡VTCNet，VRGNet與VTCNet共享部分卷積網絡以實現特征共享，并利用模型壓縮的方式壓縮模型體積，提升檢測效率，從計算機視覺和系統工程的角度對入侵輸電場景的大型機械進行精確預警。針對訓練數據偏少的問題，利用多種數據增強技術相結合的方式對數據集進行擴充。通過充分的試驗對本方法的多個超參數進行探究，綜合檢測準確率和推理速度來研究其最優配置。研究結果表明，隨著網格數目的增加，準確率也隨之增加，而召回率有先增加后降低的趨勢，檢測效率則隨著網格的增加迅速降低。綜合檢測準確率與推理速度，確定9×9為最優網格劃分方案；隨著輸入圖像尺寸的增加，檢測準確率穩步上升而檢測效率逐漸下降，綜合檢測準確率和效率，選擇480×480像素作為最終的圖像輸入尺寸。輸入實驗以及現場部署表明，相對于其他的輕量級目標檢測算法，該方法對輸電現場入侵的吊車等大型機械的檢測具有更優秀的準確性和效率，滿足實際應用的需要。Abstract: The operation of cranes and other large machinery threatens the safety of transmission lines. In order to solve this problem in the transmission scenario, the research from the aspects of data enhancement, network structure and the hyperparameters of the algorithm were performed. And a new end-to-end transmission line threat detection method based on TATLNet were proposed in this paper, which included the suspicious areas generation network VRGNet and threat discrimination network VTCNet. VRGNet and VTCNet share part of the convolution network for feature sharing and we used the model compression to compress the model volume and improved the detection efficiency. The method can realize accurate detection of large-scale machinery invading in the transmission scene from the perspective of computer vision and system engineering. To mend the insufficient training data, the data set was expanded by a combination of various data enhancement techniques. The sufficient experiments were carried out to explore the multiple hyperparameters of this method, and its optimal configuration was studied by synthesizing detection accuracy and inference speed. The research results are sufficient. With increase in the number of grids, the accuracy and recall first increase and then decrease, whereas, the detection efficiency decreases rapidly with increase in the number of grids. Considering the detection accuracy and reasoning speed, 9 × 9 is the optimal division strategy. With the increase in the input image resolution, the detection accuracy increases steadily and detection efficiency decreases gradually. To balance the detection accuracy and inference efficiency, 480 × 480 is selected as the final image input resolution. Experimental results and field deployment demonstrate that compared with other lightweight object detection algorithms, this method has better accuracy and efficiency in large-scale machinery invasion detection such as cranes in transmission fields, and meets the demands of practical applications.
-
Key words:
- deep learning /
- threat detection /
- feature sharing /
- transmission scene /
- lightweight network
-
圖 2 數據增強圖像。（a） GAN生成圖像；（b）椒鹽噪聲圖像
Figure 2. Images from data enhancement: (a)image generated from GAN; (b) image with salt and pepper noise
表 1 VRGNet中網格劃分對檢測結果的影響
Table 1. Different strategies of grid cells partitioning
Grids Precision/% Recall/% Efficiency/ms 2×2 72.23 68.49 33.61 3×3 84.80 71.99 35.85 4×4 89.60 79.59 36.48 5×5 84.37 83.87 40.37 6×6 88.48 86.90 45.62 8×8 92.62 90.14 47.66 9×9 95.19 92.40 51.63 10×10 93.28 95.15 67.21 12×12 81.14 84.36 8.29 14×14 75.61 84.49 7.29 15×15 75.11 86.30 6.05 
下載: 導出CSV
表 2 數據增強效果
Table 2. Effect of data enhancement
% Data enhancement methods Precision Recall Original images 78.19 71.52 Traditional methods 85.73 81.35 GAN 93.62 90.55 GAN+traditional methos 95.19 92.40
下載: 導出CSV
表 3 不同輸入圖像尺寸的比較
Table 3. Comparison of different image scales
Image scales Precision/% Recall/% Efficiency/ms 240×240 64.71 59.32 30.75 320×320 68.55 64.08 39.65 416×416 80.24 81.46 47.39 480×480 95.19 92.40 51.63 640×640 92.10 95.14 185.19 960×960 95.14 95.72 486.49
下載: 導出CSV
表 4 與其他方法的比較
Table 4. Comparison with other methods
Methods Precision/% Recall/% Efficiency/ms TATLNet 94.68 92.40 51.63 MobileNet 88.35 82.47 67.48 ShuffleNet 83.65 84.91 58.78 Uncompressed TATLNet 95.19 93.15 253.64
下載: 導出CSV
中文字幕在线观看表 5 現場部署檢測統計
Table 5. Detection statistics in field deployment
Alarms Actual number of intrusions Correct alarms Precision/% Recall/% Efficiency/ms 79 76 74 93.67 97.37 96.10
下載: 導出CSV
-
參考文獻
[1] Minker G A. Transmission Line Safety Monitoring System: U.S. Patent, 6377184. 2002-4-23 [2] Luo X, Zhang L Y, Luo W J, et al. Research on UAV patrol control system based on pyroelectric infrared sensor. Technol Econom Guide, 2019, 27(8): 3羅霞, 張良勇, 羅文金, 等. 基于熱釋電紅外傳感器的無人機巡檢控制系統研究. 科技經濟導刊, 2019, 27(8):3 [3] Lu Y X, Kumar A, Zhai S F, et al. Fully-adaptive feature sharing in multi-task networks with applications in person attribute classification // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, 2017: 5334 [4] He Y H, Zhang X Y, Sun J. Channel pruning for accelerating very deep neural networks // Proceedings of the IEEE International Conference on Computer Vision. Venice, 2017: 1389 [5] Han S, Mao H Z, Dally W J. Deep compression: compressing deep neural networks with pruning, trained quantization and huffman coding[J/OL]. arXiv preprint (2016-02-15)[2019-09-15]. https://arxiv.org/abs/1510.00149 [6] Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets // Advances in Neural Information Processing Systems. Montreal, 2014: 2672 [7] Duan S Z, Wei K Q. Research on active early warning monitoring system for preventing external force damage of transmission lines. Theor Res Urban Constr, 2017(15): 6段樹忠, 魏可強. 輸電線路主動預警式防外力破壞監控系統研究. 城市建設理論研究, 2017(15):6 [8] Guo S, Zeng Y H, Zhang J B, et al. Application of intelligent monitoring system for external force damage prevention for transmission lines. Guangdong Electr Power, 2018, 31(4): 139郭圣, 曾懿輝, 張紀賓, 等. 輸電線路防外力破壞智能監控系統的應用. 廣東電力, 2018, 31(4):139 [9] LeCun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553): 436 doi: 10.1038/nature14539 [10] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks // Advances in Neural Information Processing Systems. Lake Tahoe, 2012: 1097 [11] Papageorgiou C P, Oren M, Poggio T. A general framework for object detection // Sixth International Conference on Computer Vision. Tampa, 1998: 555 [12] Jiao L, Zhang F, Liu F, et al. A survey of deep learning-based object detection. IEEE Access, 2019(7): 128837 [13] Zou Z, Shi Z, Guo Y, et al Object detection in 20 years: a survey[J/OL]. arXiv preprint (2019-05-13)[2019-09-15]. https://arxiv.org/abs/1905.05241 [14] Liu W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector // European Conference on Computer Vision. Amsterdam, 2016: 21 [15] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Amsterdam, 2016: 779 [16] Law H, Deng J. CornerNet: detecting objects as paired keypoints // Proceedings of the European Conference on Computer Vision. Munich, 2018: 734 [17] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Columbus, 2014: 580 [18] Girshick R. Fast R-CNN // Proceedings of the IEEE International Conference on Computer Vision. Santiago, 2015: 1440 [19] Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks // Advances in Neural Information Processing Systems. Montreal, 2015: 91 [20] He K M, Gkioxari G, Dollár P, et al. Mask R-CNN // Proceedings of the IEEE International Conference on Computer Vision. Honolulu, 2017: 2961 [21] Huang R, Pedoeem J, Chen C X. YOLO-LITE: a real-time object detection algorithm optimized for non-GPU computers // 2018 IEEE International Conference on Big Data (Big Data). Seattle, 2018: 2503 [22] Howard A G, Zhu M, Chen B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[J/OL]. arXiv preprint (2017-04-17)[2019-09-15]. https://arxiv.org/abs/1704.04861 [23] Zhang X Y, Zhou X Y, Lin M X, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, 2018: 6848 [24] Radford A, Metz L, Chintala S. Unsupervised representation learning with deep convolutional generative adversarial networks[J/OL]. arXiv preprint (2016-01-07)[2019-09-15]. https://arxiv.org/abs/1511.06434 [25] Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, 2017: 2117 -
點擊查看大圖
圖(6) / 表(5)
計量
- 文章訪問數: 1716
- HTML全文瀏覽量: 1014
- PDF下載量: 24
- 被引次數: 0
