An Automatic Marking Method Based on Object Detection Algorithm in Aerial Images

Abstract

Aerial images of brassica napus in field were taken by UAV equipment but often contained unnecessary areas, which caused interferences and inconveniences to the subsequent study of identification of flowering stage by brassica napus images. To solve above problems, YOLO-V4 object detection algorithm based on Convolutional Neural Network (CNN) is adopted to build a model to detect brassica napus distribution area in aerial images, and model is trained iteratively by training data sets and setting initial parameters. As a result, the mean average precision (mAP) of the object detection model reached 96.24%, and the average loss is only 0.5843. In addition, the generalization and robustness ability of model have been evaluated by testing data sets. Conclusion shows the model proposed in this paper can indeed achieve an application of object detection and automatic labeling in aerial brassica napus images with a good performance.