The Detection of Engineering Vehicle on Remote Sensing Image Using Improved YOLOv5

Construction vehicles play an important role on construction sites. Real-time and accurate monitoring of these vehicles can help oversee site activities, improve work efficiency, and enhance safety. A novel engineering vehicle object detection algorithm based on an enhanced YOLOv5 is proposed since the challenges faced in object detection of engineering vehicles in remote sensing images such as complex backgrounds and low detection accuracy. To improve the detection accuracy of the model, three modules are introduced. First, the Backbone’s CSP Bottleneck with 3 convolutions (C3) modules are replaced with the CSP Bottleneck with 2 convolutions (C2f) modules to enhance the accuracy and robustness of the improved model. Next, Global Attention Mechanism (GAM) is added after the fourth C2f in the Backbone to precisely capture crucial features. Finally, the Neck’s fixed convolutions are substituted with the efficient and lightweight Group Shuffle Convolution (GSConv), which ensures a more lightweight improved model without compromising accuracy. Experimental results demonstrate that compared to the YOLOv5 and some other benchmark frameworks, our enhanced algorithm shows improvements in precision, recall and F1-score, which indicates superior performance in engineering vehicle detection.