Image Classification Based on L-M Algorithm to Improve BP Neural Network

This study mainly explores the application effect of L-M algorithm (Levenberg-Marquardt algorithm) in improving BP neural network image classification task. Although there are some precedents for the application of L-M algorithm to improve BP neural network, the difference of this study is that we optimized it specifically for MNIST dataset and adjusted the model parameters by adaptive method. This study uses the MNIST dataset, which contains a large number of handwritten digit images and provides a rich sample for image classification. By preprocessing and normalizing the data set, the BP neural network model based on L-M algorithm is constructed and trained for several rounds. The experimental results show that the improved model has achieved significant improvement in accuracy, precision, recall and F1 score, among which the accuracy is increased by about 10% compared with the traditional BP neural network, and the F1 score is also increased by nearly 9 percentage points. Especially in the later stage of training, the loss value of the model converges to a relatively stable value, indicating that the model has reached a relatively optimal parameter state. The L-M algorithm performs excellently in the task of image classification of the improved BP neural network, and effectively improves the training efficiency and classification performance of the neural network. The research results provide new ideas and methods for the field of image classification, and have certain theoretical significance and practical application value. In the future, we will continue to explore more advanced algorithms and technologies to further improve the accuracy and efficiency of image classification.