Infrared thermal imaging defect detection method and apparatus for substation insulator

An infrared thermal imaging defect detection apparatus for a substation insulator, comprising an infrared module and a core computing module, wherein the infrared module is bidirectionally connected to the core computing module, and an infrared module is unidirectionally connected to a high-rate battery;

• the core computing module is unidirectionally connected to a touch screen;
• and the infrared module is configured to obtain infrared radiation information through a lens, and calculate and generate a temperature distribution video;
• wherein an improved You Only Look Once version 5 (Yolov5) algorithm model is used in infrared image identification of the substation insulator, and a computing process is performed by using a graphics processing unit (GPU) of the core computing module;
• a defect diagnosis is implemented based on an adaptive feature temperature extraction algorithm, and is jointly implemented by using a central processing unit (CPU) and the GPU of the core computing module, and a specific defect diagnosis process is that a compiled linux executable file is run in the core computing module, the file is loaded with a trained improved Yolov5 algorithm weight file, the GPU performs a convolution operation on a photographed video stream or picture, and finally inputs a video stream or picture obtained after the convolution operation into a classifier and preselected boxes Anchor boxes identify the substation insulator, and after the substation insulator is detected, the substation insulator defect is diagnosed based on the adaptive feature temperature extraction algorithm and a defect level judgment criterion;
• and a touch display screen displays infrared thermal images, identified substation insulators, and defect levels in real time;
• wherein before the convolution operation, the GPU converts original red, green, blue (RGB) three channels of the photographed video stream or picture into a grayscale image single channel, wherein the converted grayscale image single channel is inputted the improved Yolov5 algorithm model;
• and
• wherein during the convolution operation, an image is sliced through a Focus structure in a backbone section and input to a subsequent Cross Stage Partial (CSP) structure, wherein a Neck part uses an Feature Pyramid Network and Path Aggregation Network (FPN+PAN) structure and uses a Cross Stage Partial version 2 (CSP2) structure designed by a Cross Stage Partial Network (CSPnet) for reference to enhance a capability of feature fusion.