Improved YOLOv8 Remote Sensing Small Target Detection Method

Abstract

Due to the miniaturization, dense arrangement, variable viewing Angle and complexity of background environment in satellite remote sensing technology, the traditional detection methods often encounter the challenge of misidentification and missing detection, thus limiting the detection accuracy. In order to overcome these problems and improve detection efficiency, this paper innovatively proposes an optimized Yolov8 model, which is deeply customized and improved for the detection of tiny objects in remote sensing images. Firstly, BiFormer model is introduced. BiFormer model introduces a two-layer routing attention mechanism, which significantly improves the accuracy and robustness of target detection. In addition, the SPD-Conv module is introduced to realize the conversion from space to depth, and better capture the target features of different dimensions. After rigorous validation of the DOTA-v1.0 dataset, the optimized model achieved a significant improvement in average accuracy (mAP), reaching a level of excellence of 60.3%, which represents a performance leap of about 2.3 percentage points compared to traditional models. It has further promoted the technological progress and application deepening in this field.