Design of Remote Sensing Image Target Detection Algorithm based on Machine Vision

Aiming at the problem of low detection accuracy in current remote sensing image target detection due to factors such as large target size differences, complex backgrounds, and imbalanced categories, this paper proposes an enhanced YOLOv8s model based on machine vision technology. First, in order to minimize the amount of computation while ensuring high detection accuracy, a GhostNet convolutional module is added to the backbone network; then, the EMA channel and spatial attention mechanism are combined to enhance the model’s ability to extract important features and effectively suppress the interference of complex backgrounds; secondly, the neck network design adopts a bidirectional feature pyramid network, and the top-down and bottom-up multi-scale feature fusion further enhances the model’s multi-scale target recognition ability. The proposed enhanced model outperforms other comparison models, and the mAP@0.5 value on the public dataset DIOR is 3.1% higher than the baseline model, achieving a high detection accuracy of 71.9%. The strategy proposed in this paper provides an efficient and robust solution for remote sensing image target detection, which has strong theoretical value and practical application prospects.