Efficient Object Detection in Remote Sensing Images Based on Feature Weaving and Redundancy Suppression

Remote sensing object detection faces significant challenges in handling scale variations, complex backgrounds and dense target distributions. Existing models often struggle with high computational costs and massive number of parameters, hindering deployment on resource-limited devices. To address these issues, we propose an efficient object detection network based on feature weaving and redundancy suppression called EWS-YOLO, featuring two components: (1) The Weave-Gated Parallel (WGP) module employs grouped convolution with interlaced cascading and gating mechanisms to achieve implicit multi-scale feature extraction and adaptive fusion, significantly enhancing multi-scale target detection capability. (2) The Redundancy-Suppressed Feature Pyramid Network (RFS-FPN) integrates an Progressive Shuffle Upsampling mechanism coupled with Dynamic Feature Fusion, enabling intelligent fusion of low-level spatial details and high-level semantic features while effectively suppressing redundant information through adaptive channel-wise feature recalibration. Experimental results on VisDrone, DIOR and UAVDT datasets demonstrate EWS-YOLO's superior performance, achieving 42.4%, 95.2%, 98.7% mAP@.50 respectively while reducing parameters by 41% compared to YOLOv8s, making it particularly suitable for resource-constrained applications.