Research on the optimization method of image classification model based on deep learning technology and its improvement of data processing efficiency

Deep neural networks, as an outstanding representative of the field of artificial intelligence, have been widely used in various fields, and have shown performance beyond human in image classification tasks. This study is based on deep learning technology to explore the optimization path of image classification model, this paper uses particle swarm algorithm for classification optimization, on this basis, considering the long-tailed distribution of real image data samples, this paper, on the basis of Softmax cross entropy loss research, combined with double angle sinusoidal attenuation strategy to integrate the BACL and NCE loss in order to build a new joint training framework, so as to improve the classification performance of the classification performance of the long-tailed classification model, a data processing method based on sample gradient optimization is proposed. In the model performance experiments, the accuracies of VGG16, ResNet18, MobileNetV2, and InceptionV3 were improved by adding deep information data, which improved by 4.2%, 2.6%, 1.6%, and 3.1%, respectively. And the improved network model in this paper has the smallest loss, which basically stays around 0.10. In addition, the weighted accuracy of this paper’s model reaches 98.23%, which has a better classification and recognition effect compared to several other networks. On the other hand, the training time of this paper’s model is only 135 minutes, which saves about double the time compared to other models. The model in this paper identifies and classifies seven types of life images, and the classification correct rate is higher than 85%, and the overall classification performance is excellent, and the results show that the image classification optimization model based on deep learning in this paper has excellent performance and has certain practical application effects.