Information Extraction of the Vehicle from High-Resolution Remote Sensing Image Based on Convolution Neural Network

Abstract

Aims: To effectively detect vehicle targets in remote sensing images, it can be widely used in traffic management, route planning, and vehicle flow detection. YOLOv3 deep learning neural network, which mainly studies the vehicle target detection in remote sensing images and carries out the target detection suitable for the characteristics of remote sensing images. Objective: This paper studies the information extraction of vehicle high-resolution remote sensing images based on a convolution neural network. Method: The YOLOv3 network model of vehicle target detection in satellite remote sensing images is optimized. The iterations are set to 50002000045000, and the learning rate is 0.001. At the same time, the comparative experiments of RCNN, Fast RCNN, fast RCNN, and yolov3 network models are carried out. Result: The ca-yolov3 network model can be applied to target detection in satellite images. After 40500 times of learning, the loss function value of the model is reduced to about 0.011. Conclusion: The IOU value of the model also has a good performance in the training process, which makes the yolov3 neural network model more accurate in the image small target detection.