Fine-Grained Object Recognition Algorithm Based on Self-Supervised Representation Learning of Multiview Infrared Images

Utilizing multi-view infrared images to collaboratively identify the types of surface ship targets is a feasible approach in practice. This paper proposes a fine-grained object recognition algorithm based on self-supervised representation learning of multi- view infrared images. A contrastive learning framework for target recognition based on multi-view infrared images is introduced, employing a training method of pre-training with unlabeled data and fine-tuning with a small amount of labeled data, which enhances the generalization ability of the algorithm. By optimizing the instance construction method, the similarity and differences between instances are better extracted, learning invariant features of the same target from different views; by combining the cross-correlation matrix, redundancy in feature representation is reduced; a target recognition application method with adaptive input of multi-view images is proposed, improving the accuracy of downstream target recognition tasks; a multi-view infrared image dataset of ship targets is constructed and algorithm validation is performed. Experimental results show the Accuracy, Precision, Recall, and F1 score of proposed method are 95.1%, 95.2%, 95.3%, and 95.2%, respectively, which are higher than the results of the SimCLR algorithm by 7.1%, 6.7%, 6.8%, and 6.7%, and higher than the results of the BYOL algorithm by 7.7%, 7.6%, 7.3%, and 7.4%.