| Summary: | This paper builds a power operation target detection model based on the YOLOv4 algorithm in intelligent image recognition, and optimizes the YOLOv4 algorithm by combining with the loss function to improve the accuracy of power target operation detection. The kmeans++ algorithm was used to cluster the electric power operation behaviors to obtain a more accurate electric power operation behavior dataset. Three sets of tests were conducted after the model was constructed, targeting the behavioral set of electric power workers in a certain place and the behavior in VOC format, followed by the multi-target tracking effect test. The analysis based on the obtained data showed that the helmet placement detection confidence, fatigue detection confidence, smoking detection confidence, and fall detection confidence reached 0.97, 0.93, 0.89, and 0.93, respectively. The transmission speed got 53.58 fps, and the recall and precision of the multi-target tracking were also above 93%. The YOLOv4 detection model based on keans++ clustering algorithm can effectively detect and identify the variable power operation behavior images.
|
|---|