Lu Wang, Dong Ren, Xiaoran Tian, Yunjie Zhang, and Deao Hu
Pine wilt disease, double-head, sampling strategy, UAV
The detection method of pine wilt disease tree using UAV images is effective. When using the deep learning method to detect pine wilt disease, the problem that the number of diseased trees is so small compared with non-diseased trees that there is an imbalance between the positive and negative samples. This paper proposes a sampling threshold interval weighting method. The sampler is redesigned, and the IOU interval is divided into batches. The interval of difficult samples is weighted to improve the sampling rate of difficult samples and suppress the simple samples’, so as to avoid sampling a large number of easily negative samples in random sampling, which leads to the imbalance of sample difficulty. In addition, a double-head detection is introduced to resolve the problem, that is, the spatial misalignment caused by the common feature parameters of classification and regression in the traditional R-CNN network. To verify the effectiveness of the proposed method, we conducted ablation experiments and comparative experiments on UAV images taken from the city of Yichang. The experimental results showed that the F1 accuracy of the improved Faster R-CNN network reached 91.46%, which could achieve accurate detection of pine wilt disease tree.
Important Links:
Go Back
