Image investigation has become one of the main methods of military investigation. Because of the large amount of data detected in the investigation, how to classify images correctly in the early stage and improve the efficiency of image processing in the later stage has become the focus of research. The features reflected in the image information of different target categories are different. Image classification refers to the distinction of different target categories by features. A feature can not describe the information of an image comprehensively. It combines texture features and gray statistics features into comprehensive features. The multi-layer perceptron has a remarkable ability of learning and reasoning, and it can solve the problem of complex classification. Therefore, a method of image classification based on the combination of image features and multi-layer perceptions is proposed. An image classification system is designed and implemented, using standard image library to conduct experiments. Firstly, the texture features and grayscale features are extracted, and then the selected eigenvalues are combined into eigenvectors for normalization, which is used as the input of the multi-layer perceptron, and the predicted image classes are used as the output of the multi-layer perceptron, and the result of classification is obtained. The classification accuracy is more than 0.8, and the classification system is applied to the test result evaluation system of a certain type of machine. The classification effect is good, which can provide reference for the related application of image processing system.