Fig. 3

Primary diagnosis unit for bounding box detection. Figure 3 shows the process by which the primary diagnostic unit detects suspicious areas containing disease and pests in the input image. It is similar to Fuentes et al. (2017b). Compared the manual setting of boundary boxes with the k-means++ clustering generation of boundary boxes, in which the red solid border is the correct border and the red dotted border is the highest border in IoU of different methods of generating boundary boxes. a The size and shape of the default boundary box set manually are usually not able to obtain high IoU, and b the prior knowledge of the boundary box is obtained from tens of millions of samples using k-means++ (thin, high borders account for the majority)