CSPPartial-YOLO: A Lightweight YOLO-Based Method for Typical Objects Detection in Remote Sensing Images

Detecting and recognizing objects are crucial steps in interpreting remote sensing images. At present, deep learning methods are predominantly employed for detecting objects in remote sensing images, necessitating a significant number of floating-point computations. However, low computing power and small storage in computing devices are hard to afford the large model parameter quantity and high computing complexity. To address these constraints, this article presents a lightweight detection model called CSPPartial-YOLO. This model introduces the partial hybrid dilated convolution (PHDC) Block module that combines hybrid dilated convolutions and partial convolutions to increase the receptive field at a low computational cost. By using the PHDC Block within the model design framework of cross-stage partial connection, we construct CSPPartialStage that reduces computational burden without compromising accuracy. Coordinate attention module is also employed in CSPPartialStage to aggregate position information and improve the detection of small objects with complex distributions in remote sensing images. A backbone and neck are developed with CSPPartialStage, and the rotation head of the PPYOLOE-R model adapts to objects of multiple orientations in remote sensing images. Empirical experiments using the dataset for object deTection in aerial images (DOTA) dataset and a large-scale small object detection dAtaset (SODA-A) dataset indicate that our method is faster and resource efficient than the baseline model (PPYOLOE-R), while achieving higher accuracy. Furthermore, comparisons with current state-of-the-art YOLO series detectors show our proposed model's competitiveness in terms of speed and accuracy. Moreover, compared to mainstream lightweight networks, our model exhibits better hardware adaptability, with lower inference latency and higher detection accuracy.