Analyzing Spatial Distribution Patterns of Wheat Ears

The management of wheat breeding and cultivation can be launched accurately through the assistance of the visualization of wheat ears counting and spatial distribution. This study aimed to develop an automatic and accurate counting method, and a visualization way of spatial distribution of wheat ears from it. Based on unmanned aerial vehicle (UAV) acquisition and Global Wheat Head Detection (GWHD) dataset, the wheat ears dataset was constructed. Five typical deep learning target detection models, including Faster R-CNN, SSD, RetinaNet, YOLOv3 SPP and YOLOv5s, were selected for training, verification, and testing. Among them, Faster R-CNN and YOLOv5s models showed the most satisfying counting effect, with the accuracy of wheat ears recognition above 95%. After the improvement of the two models, the ears recognition accuracy of Faster R-CNN model exceeded 96%, while the ears recognition accuracy of YOLOv5s model exceeded 98%. In order to provide decision-making basis for the precise management of wheat farmland, this study explored and realized the density and size distribution visualization of wheat ears by using image processing technology. In addition, the shape distribution visualization of wheat ears was studied and accomplished with the foundation of U-net model. In particular, the number distribution visualization of wheat ears in the whole target wheat field in video data was performed effectively by adopting the combination of YOLOv5s+DeepSort and Python program.