基于半监督的植物病害智能检测研究OA

Food security has always been a matter related to the people's livelihood.In order to increase food production,a large amount of pesticides are overused,posing a threat to the environment.Because the characteristics of tea diseases are relatively obvious and easy to annotate,so the tea is taken as the research target,hoping to use semi-supervised deep learning methods to help tea garden managers quickly find areas with relatively severe tea diseases in the tea garden,so as to promote the efficient use of pesticides and reduce the pollution of pesticides on the environment.Although the fully supervised method can achieve this function well,it requires a huge amount of data,which undoubtedly increases the cost.Therefore,a semi-supervised method is adopted to achieve this function,achieving the goal of reducing costs while maintaining good performance.The study first used drones to conduct low-altitude aerial photography of designated tea gardens.Then,specific frames are extracted,and relevant frames are manually annotated.Finally,the annotated dataset is further processed to obtain a standard COCO format dataset.Fully supervised and end to end semi-supervised experiments are conducted on the dataset to verify the effectiveness of the semi-supervised framework used.