SSAC-YOLO: An Adaptive Target Detection Algorithm for Weak Small Objects in Remote Sensing Images

Remote sensing images are widely applied in land resource surveys, urban planning, and related fields, but traditional detection algorithms often suffer from high miss rates and inadequate feature extraction due to challenges such as densely distributed small targets and complex background interference. To address these limitations, this study proposes SSAC-YOLO, an adaptive target detection algorithm for remote sensing small targets, which significantly enhances detection performance through comprehensive optimization of YOLO. The algorithm introduces three key innovations: (1) a novel multi-scale feature pyramid structure combining cross-stage feature fusion and multi-scale pooling to improve detection accuracy across varying target sizes; (2) a dynamic switchable atrous convolution (SAC) module that adaptively adjusts receptive fields and integrates global contextual information to enhance small-target feature representation; and (3) a hybrid data augmentation strategy incorporating adversarial perturbation simulation to boost robustness in complex scenes. Our experimental validation on the RSOD, SIMD, and SODA benchmarks confirms that SSAC-YOLO achieves better results than leading approaches in major evaluation criteria such as mAP50, particularly excelling in dense small-target scenarios. With its balanced efficiency and lightweight design, the framework provides an advanced solution for high-resolution remote sensing image analysis.