Aiming at the problems of image noise pollution and weak distinguishability of defect features under complex working conditions in industrial intelligent maintenance, this study proposes an intelligent maintenance image model based on the Multi-Objective Differential Evolution Algorithm (MODE). The model extracts macro-to-micro defect representations through multi-feature fusion of the Gray-Level Co-Occurrence Matrix (GLCM) and Local Binary Pattern (LBP), uses the Multi-Objective Differential Evolution Algorithm (MODE) to collaboratively optimize the parameters of the Convolutional Neural Network (CNN) to balance model accuracy and complexity, and enhances industrial scene adaptability by combining multi-scale feature networks with physics model-driven noise suppression strategies. Experiments show that the model achieves a crack classification accuracy of 91.7%, improves anti-interference capabilities against Gaussian noise, motion blur, etc., by 16.6%-22.9% compared with traditional methods, and has an end-to-end detection delay of 153.3 ms, providing an efficient solution for real-time intelligent maintenance of industrial equipment.