无人机核查广西重点松材线虫疫情的应用实践

Application and Practice of UAV in Checkingthe Key Epidemic Situation of Pinewood Nematodes in Guangxi

将包含枯死松木的松树林历史无人机影像输入到U-net神经网络中进行学习,以人工标注样本作为验证,得到识别准确率为83.44%,漏检率为12.61%,误检率为11.87%的自动识别模型,以该模型对贺州市平桂区羊头镇、八步区桂岭镇和玉林市博白县那林镇3个重点松材线虫疫区,共18.39 km^(2)区域的松树林无人机影像进行自动识别并辅以人工修正,得到最终识别成果。共发现枯立木5103株,清除痕迹148个,枯倒木52株,涉及林班19个,小班470个,面积911.75 hm^(2)。应用模型对大范围无人机影像进行识别,辅以人工修正,相较于单纯的人工识别,准确率相当,却能够省时省力,极大地提高工作效率。值得大范围松材线虫疫情核查推广应用。

The historical UAV images ondead pine trees were input into the U-net neural network for learning,and the manually labeled samples were used as verification.The automatic recognition model with the recognition accuracy rate of 83.44%,the missed detection rate of 12.61%,and the false detection rate of 11.87%was obtained.The model was used to identify three major pine wood nematode epidemic areas,including Yangtou Town in Pinggui District of Hezhou City,Guiling Town of Babu District,and Nalin Town in Bobai County of Yulin City.The UAV images in a total area of 18.39 km^(2) are automatically identified and supplemented by manual correction to obtain the final recognition result.Results obtained as follows:a total of 5103 dead pine trees,148 traces,and 52 dead fallen trees were found,involving 19 compartment and 470 subcompartment,covering an area of 911.75hm^(2).The application of the model to the recognition of largescale UAV images,supplemented by manual correction,has the same accuracy but can save time and effort,which greatly improves the recognition efficiency,and provides reference for the verification of largescale pine wood nematode.