Abstract
In order to improve the accuracy of induction motor state detection, aiming at the problems of low resolution, poor contrast, and insufficient detail in infrared thermal image recognition, an Adaptive Texture Bi-histogram Equalization (ATBHE) is proposed. ATBHE's image enhancement method improves images' visual effects and highlights detailed information. Firstly, the histogram is preprocessed by the moving average filtering method, and suitable minimum points are found as segmentation points to ensure correct pixel distribution. Then, an adaptive threshold platform is constructed according to the mean value of the sub-histogram, and the clipping threshold is constructed according to the Canny edge detection operator of the image and the related texture characteristics of GLCM. Finally, the sub-histograms are combined into enhanced images after histogram equalization to solve the problem of over-enhancement and under-enhancement. Through experiments and objective evaluation such as SSIM, PSNR, information entropy, and average gradient, the enhanced image quality and visual effect of the proposed method are better. At the same time, the infrared thermal imaging induction motor image enhanced by this method is applied to the recognition of convolutional neural networks, and the induction motor state detection can be realized well.