Utilization of similar detection dataset based on dual classification head

To address the challenge of insufficient accurately labeled data in specific object detection tasks, this paper introduces a dual classification head approach that effectively leverages similar datasets. The proposed method trains data from diverse datasets through distinct classification heads, with similar data actively participating throughout the entire training process. This innovative approach establishes a novel paradigm for bridging data gaps and enhancing model accuracy. Experimental evaluations conducted on the fine-grained object detection competition dataset and the FAIR1M dataset demonstrate that the proposed method substantially improves the accuracy and robustness of object detectors in remote sensing images.The code will be available soon at https://github.com/zf020114/DCH .