Wheat seedling line detection is critical for precision agriculture and automatic guidance in early wheat field operation. Aiming at the complex wheat field environment, a method of detecting wheat seedling lines base...Wheat seedling line detection is critical for precision agriculture and automatic guidance in early wheat field operation. Aiming at the complex wheat field environment, a method of detecting wheat seedling lines based on deep learning was proposed in this study. Firstly, a rotated bounding box was created to improve the YOLOv3 model to predict the approximate position of the wheat seedling line;Then, according to the rotated bounding region obtained by the model, the wheat seedling line was detected by fitting the extracted center points. Finally, a comprehensive evaluation method combining angle error and distance error was proposed to evaluate the accuracy of the extracted crop line. By testing images of wheat seedlings in different environments, the results showed that the mean angle error and distance error respectively reached 0.75° and 10.84 pixels while the mean running time was 63.83 ms for a 1920×1080 pixels image. And compared to the original model the improved algorithm model improved the mAP value by 13.2%. The angle error and the distance error of the improved algorithm model were reduced by 51.4% and 39.7%, respectively. The method proposed in this study can accurately detect the wheat seedling lines at different stages and it is also suitable for the environments with weeds, shadow, bright light, and dark light. At the same time, it has a certain adaptability to wheat seedling images with a yaw angle in the shooting process. The research results could provide a reference for the automatic guidance of early wheat field machinery.展开更多
基金funded by the Natural Science Foundation of Shandong Province (Grant No. ZR2021MF096)Shandong Agricultural Machinery Equipment Research and Development Innovation Plan (Grant No. 2018YF009).
文摘Wheat seedling line detection is critical for precision agriculture and automatic guidance in early wheat field operation. Aiming at the complex wheat field environment, a method of detecting wheat seedling lines based on deep learning was proposed in this study. Firstly, a rotated bounding box was created to improve the YOLOv3 model to predict the approximate position of the wheat seedling line;Then, according to the rotated bounding region obtained by the model, the wheat seedling line was detected by fitting the extracted center points. Finally, a comprehensive evaluation method combining angle error and distance error was proposed to evaluate the accuracy of the extracted crop line. By testing images of wheat seedlings in different environments, the results showed that the mean angle error and distance error respectively reached 0.75° and 10.84 pixels while the mean running time was 63.83 ms for a 1920×1080 pixels image. And compared to the original model the improved algorithm model improved the mAP value by 13.2%. The angle error and the distance error of the improved algorithm model were reduced by 51.4% and 39.7%, respectively. The method proposed in this study can accurately detect the wheat seedling lines at different stages and it is also suitable for the environments with weeds, shadow, bright light, and dark light. At the same time, it has a certain adaptability to wheat seedling images with a yaw angle in the shooting process. The research results could provide a reference for the automatic guidance of early wheat field machinery.
