Mechanical structures may exhibit defects during long-time high-temperature operation. Infrared image technology quickly and automatically detects mechanical structure defects, improves detection speed and accuracy, and reduces the workload of manual detection. Although high-temperature infrared image processing method has made significant progress in detecting the defects, it still has some shortcomings. Defect features in high-temperature infrared images may not be obvious and are mixed with background information, thus making it difficult to accurately identify and extract them. Therefore, this research studied the detection method of mechanical structure defects after high temperature based on image processing. Transform domain denoising method was used to decompose the transform domains of images, which distinguished signals and noises in the images. Adaptive Contrast Enhancement (ACE) algorithm was used to enhance the images. A feature fusion imaging detection framework for infrared and optical imaging of high-temperature mechanical structures was constructed, which improved the accuracy and reliability of defect detection. Deep neural network was combined with the heuristic fusion section, which further explored deep features in the images and improved the fusion effects. The proposed fusion features were processed using binary tree classification and hierarchical classifier, which accurately identified the abnormal defect regions in the infrared images of high-temperature mechanical structures. The experimental results verified that the proposed method was effective.