A Multitask Collaborative Learning Approach for Fine-Grained Object Detection in Remote Sensing Imagery

Fine-grained object detection in remote sensing images is challenged by high intra-class variation and low inter-class distinction, as well as optimization inconsistency between classification and localization tasks. To address these, we propose MCLNet, a Multi-Task Collaborative Learning Network. MCLNet is a dual-stream architecture that integrates fine-grained classification with coarse-grained detection to improve detection accuracy. To enhance oriented localization, we treat all fine-grained objects as coarse-grained ones during training, which expands the dataset and improves the accuracy of oriented localization in the coarse-grained detection stream. Besides, we introduce a Coarse-to-Fine Hierarchical Constraint Mechanism (CHCM) that uses coarse-grained predictions as priors to guide fine-grained learning, enhancing both intra-class and inter-class feature representations. Additionally, we employ a Progressive Multi-Task Optimization Strategy (PMOS) that gradually trains the model to balance classification and localization tasks, reducing task interference. Extensive experiments on widely used benchmark datasets validate that MCLNet outperforms current state-of-the-art methods in fine-grained object detection.