Research on Multi-Labels Image Classification Based on Self-Supervised Model

At present, image classification model has become an essential component for detection system. However, existing identification models are primarily concentrated on the convolutional neural network or utilize transformer model to capture the features of training images, which can obtain acceptable accuracy for single objectives and is hard to dispose multiple objectives with unsatisfied optimization results. In this work, we utilize the self-supervised mechanism to dispose the optimization process for traditional convolutional neural network. Initially, we construct the contrastive learning module to generate the amount of objectives label for the training data, which is utilized to reduce the extraction loss for multiple tasks. Subsequently, the convolutional neural network is established to classify the objectives in the images. From our extensive experimental results and compared analysis with existing classification models, we can directly observe that our proposed model can achieve the multiple images classification with more than 85% identification accuracy with reasonable responding costs.