Rapid recognition technology of composite insulator infrared thermal image based on semantic segmentation algorithm

Aiming at the problem that typical defect types (core rod brittle fracture, sheath cracking, etc.) have complex features and are difficult to identify, a fast recognition technology for composite insulator infrared thermal images based on semantic segmentation algorithm is proposed. For the infrared thermal images of composite insulators, the format, size and other parameters are unified, grayscale conversion is performed, adaptive median filtering and mean filtering are used to remove noise, and then logarithmic transformation is used to enhance the contrast. With VGG16 convolutional neural network as the backbone, the fully connected layer is removed, a specific convolution strategy is adopted, and the activation function, maximum pooling layer, and feature channel attention mechanism are introduced to construct an insulator infrared thermal image segmentation model based on semantic segmentation algorithm. The preprocessed and annotated infrared thermal images are divided into training sets, etc., the pre-training model is selected, the training parameters are set, a specific learning rate strategy is adopted, hyperparameters are introduced, and the model is trained with a staged loss function to quickly identify composite insulators. The experimental results show that the research method significantly improves the recognition accuracy, accelerates the recognition speed, and greatly improves the recognition efficiency.