An infrared image detection algorithm for power equipment based on search optimization for YOLOv5

Accurately analysing features in infrared images of equipment is one of the current directions in the field of power equipment detection and identification. Because infrared images of power equipment have poor resolution, low contrast, and visual blurring problems, this work proposes the use of the squirrel search algorithm to optimize the detection strategy of YOLOv5. First, due to the shortcomings of the squirrel search algorithm, which easily falls into local optima and has a slow convergence speed, in this work, the Henon Consine Seagull search algorithm (HCSSA) is proposed; this algorithm uses Henon chaotic mapping for population initialization and optimizes the predator probability based on the cosine function to improve the algorithm's performance. Second, in the YOLOv5 model, CSP_Faster is used for feature information recognition and to reduce the computational burden, the SKNet mechanism is introduced to ensure the integrity of the image feature information, the SIoU loss function in target classification is used to obtain a better classification effect, and finally, the HCSSA algorithm is optimized for the two hyperparameters of the YOLOv5 model, which are the learning rate and the weight decay. In the simulation experiments, the recognition effect of the proposed algorithm is improved by 8.87 %, 7.67 % and 5.11 % compared with those of YOLOv3, YOLOv4, and YOLOv5, respectively, which shows that the model has a better target detection effect.