改进YOLOv7网络在遥感影像建筑垃圾识别的应用

Application of Improved YOLOv7 Network in Recognition of Construction Waste in Remote Sensing Imagery

针对现有的目标检测算法对于高分遥感影像建筑垃圾识别效率不高的问题,提出一种基于改进的YOLOv7的目标检测模型,提高建筑垃圾的检测效果。该方法使用高分二号卫星影像数据,首先用SIoU来优化模型的目标框回归,加快模型的收敛速度;然后用漏斗激活函数FReLU扩大卷积层的感受野范围,以提高模型的特征提取能力;最后使用深度可分离卷积核,在提高检测精度的同时也减小了模型的参数量。实验结果表明,改进后的YOLOv7模型相比其他模型平均精度、准确率和召回率分别提升了5.8%、6.4%和8%,具有较好的识别效果,为遥感影像建筑垃圾识别提供了可靠的方法。

An improved YOLOv7-based object detection model to enhance the efficiency of identifying construction waste in high-resolution remote sensing images is proposed in this paper.The method utilizes data from the Gaofen-2 satellite imagery.Firstly,the model’s bounding box regression is optimized using SIoU to expedite the convergence speed.Secondly,the funnel activation function FReLU is employed to expand the receptive field of the convolutional layers,thereby improving the model’s feature extraction capabilities.Finally,depth-wise separable convolutional kernels are utilized to enhance detection accuracy while reducing model parameters.Experimental results demonstrate that the improved YOLOv7 model achieves a 5.8%increase in average precision,6.4%improvement in accuracy,and 8%enhancement in recall compared with other models.It exhibits excellent recognition performance,offering a reliable approach for construction waste identification in remote sensing images.