Research On SAR-OPT Cross-modal Aircraft Target Detection Method Based On Contrastive Learning

In the field of remote sensing, the diversity of data modalities of different sensors limits the universality of target detection methods, and models suitable for specific modalities are difficult to generalize to new scenarios. In order to deal with the problem of missing samples caused by the occlusion of optical images by clouds or the difficulty of annotating SAR images, this paper proposes a cross-modal universal target detection method based on contrastive learning. This method utilizes the characteristics of optical and synthetic aperture radar (SAR) images, and designs targeted contrastive learning agent tasks - performing random probability grayscale conversion on optical images and applying pseudo-color image generation to SAR images, so as to prompt the model to extract and identify similar features from the two modalities under the guidance of the loss function. In addition, by introducing spatial attention and channel attention mechanisms, the former is used to enhance the model’s attention to the global features of the image, and the latter is used to balance the RGB three-channel information of the optical image and the information of the single channel of the SAR image. Experimental results show that in few-shot and zero-shot scenarios, this method can effectively identify targets and show good generalization ability and robustness.