Research On Image Recognition Technology For Power Operation Violations

The power industry guarantees national life as well as national security.However,the electric power operation is often in dangerous environments such as high altitude and high voltage,which is very likely to occur safety accidents,resulting in casualties and property damage.At present,the detection of power operation violations mostly focuses on the detection of a single power operation scene or a single risk type using image recognition algorithms,and lacks the work of detecting multiple types of power operation risks at the same time.In view of the above problems,this paper researches the detection of multiple types of power operation violations based on image recognition technology,and the specific work done is as follows:1.For multiple types of electric power operations that have special objects that can be used as scene recognition features,an electric power operations violation detection method based on object recognition is designed.Firstly,K-means++ algorithm is used to cluster the anchor size suitable for the electric power industry,and then specific objects are detected by YOLO v4 to determine the electric power operation scenes;and Do I(Degree of Intersection)index is defined to evaluate whether the protective equipment is correctly worn,and finally safety index is constructed to detect whether there is a violation of the operation behavior.After the verification of the "smoking" and "welding" datasets,the algorithm detects 37 fps on RTX TITAN,and the detection accuracy is 94.39%.2.For many types of electrical operations with obvious operating posture,a method to detect electrical operation violations based on personnel posture recognition is designed: The coordinates of key points of the human body are extracted by Open Pose,which is transformed into human skeleton map and transmitted to GCN(Graph Convolutional Networks)to judge the operation scenes.At the same time,introducing the attention mechanism to learn the weight of different limbs according to their importance to the gesture classification;Finally,the distance between the protective equipment to be worn and the human body part is calculated,and combines the results of posture classification to build a logical judgment function to detect violations in electrical operations.After the verification of the "smoking","working at height" and "crossing the guardrail" datasets,the algorithm detects 11 fps on RTX TITAN,and the detection accuracy is 94.42%.3.In order to detect the risk of power operation scene with both object and posture features and power operation scene have neither object and posture features,a model for detecting power operation violations based on fused object and posture features is constructed.The model can be divided into two branches: the object detection branch uses the YOLO v4 algorithm to extract object information,and the pose estimation branch uses the Open Pose-GCN network to extract the operator’s pose information,and then combines the outputs of the two branches construct fusion features,and then uses the OS-ELM(Online Sequential Extreme Learning Machine)to detect whether there are violations.Based on the validation of 7 types of power operation datasets,the model has a wider range of applicability and a higher detection accuracy of 97.31%.4.A software system applied to substation power operation risk recognition is designed and developed,and the interface is demonstrated.Verified by functional and performance tests,the detection method of power operation violation proposed in this paper is reasonable and feasible,and the system realized human-computer interaction well.The software system is successfully applied in the substations of Guangxi Power Grid.