Research On Electrical Equipment Object Detection Based On Deep Learning

Object detection based on deep learning is an important research direction in the field of artificial intelligence.But it is not good at occluded objects detection.Aiming at the domestic and international researches on object detection methods,this paper proposes an improved method to enhance the occlusion target detection,and applies the improved method to the case review of the insurance industry,so as to reduce the workload of employees and reduce the labor cost of the company.The research work includes the following contents:1.In view of the current object detection methods based on deep learning,an improved network based on Faster R-CNN is proposed,and Resnet50 residual network is introduced to enhance the feature extraction ability;FPN network is introduced to fusion the context information;The dilated convolution is introduced to enhance the ability of occlusion detection;ROI align is introduced to improve the object position accuracy of the whole network.Then,in the process of training model,the data of electrical category label are smoothed,and the cosine annealing function is used to control the learning rate in the stochastic gradient descent algorithm.The effectiveness of the improved method was verified in COCO2017,MAFA and Wider Face datasets.2.Based on the needs of studying practical problems,we put forward the selection criteria for collecting images of data sets to ensure the quality of sample data.The labeling of the target in the image is designed,and the images of different parts of the same electrical appliance are labeled into different categories,so as to facilitate the training model to learn the characteristics of different parts.Traditional methods such as rotation,translation,flip and random color jitter were used to augment the data set.Then,the data set is further amplified by using the visually coherent image mixup method in the process of training,and the exponential decay function is used to control the learning rate in the stochastic gradient descent algorithm,so as to improve the robustness of the model.3.Based on the above research work,the design and implementation of the electrical equipment object detection module in the electrical maintenance case management system are completed.This paper mainly introduces the system structure design,database interface table design,electrical test model training,data processing flow design and development content of the electrical test module,and carries on the effect demonstration of the related functional modules.The electrical equipment object detection module is applied to electrical maintenance cases scenario.