Research On Object Detection Method Of Power Distribution Equipment Based On Deep Learning

With the rapid growth of China’s national economy,the demand for electricity is rapidly increasing,and the overall power supply is tight,ensuring t he normal operation of power equipment has become a problem that cannot be ignored.Power distribution equipment is an important part of the power system,which is prone to damage and performance degradation due to long-term exposure to various environments,affecting the overall stability of the power supply in the distribution network.For power systems,achieving accurate identification of different power equipment and using it as a prerequisite for assessing the occurrence of faults in power equipment is an important measure to ensure the safe and reliable operation of power systems.In this paper,using deep learning related technology,the current target detection algorithm YOLOv3 is selected to address the shortcomings of traditional power equipment detection,and an improvement method is proposed to improve the accuracy of the model for the identification of different power distribution equipment and to realize the target detection of power distribution equipment,which is in line with the trend of the power industry to develop in the direction of intelligence and high quality.The research content of this paper is as follows.(1)To address the problems of insufficient utilization of shallow information and low detection accuracy,a DC-YOLOv3 target detection model that incorporates densely connected modules and attention mechanisms is proposed.The first two sets of residual modules in Dark Net-53 are replaced with densely connected modules to enhance the feature reuse of shallow features in network propagation.At the same time,the attention mechanism CBAM is introduced to the residual network to improve the expression of network features.Finally,the depth-separable convolution is replaced with the traditional convolution to reduce the number of mod el parameters.The model is tested on the collected distribution equipment dataset,and the detection accuracy of the algorithm is significantly improved,which verifies the effectiveness of the model.(2)In order to further improve the detection accuracy as well as the leakage and false detection problems of device detection,the Final-YOLOv3 target detection model with bi-directional feature information fusion and multi-scale detection is proposed on the basis of DC-YOLOv3.The original FPN structure is improved,top-down feature information flow is added,bidirectional fusion of shallow and deep features is realized,the extracted features are richer,and the original three scale detection is changed to four detection scales to improve the detection capa bility of the target.The model is tested on the collected power distribution equipment dataset,and the experiments show that the detection accuracy of Final-YOLOv3 algorithm is improved,and different power distribution equipment can be effectively detec ted.(3)To better match the anchor frame size of the power distribution equipment,the K-Means algorithm is used to generate new anchor frames on the collected power distribution equipment dataset.Meanwhile,GIo U is introduced to replace the original Io U in the YOLOv3 model to improve the accuracy of the target detection results and make the model converge better.