Stem-leaf angle is an important phenotypic parameter of rice, which is crucial for the growth and development of rice plants and scientific breeding. However, the traditional method is inefficient and subjective. In order to quickly and accurately calculate the stem-leaf pinch angle of rice, this paper proposes a key point-based stem-leaf pinch angle detection method. Experimentally, four periods of rice (Jointing stage, Booting stage, Heading stage and Mature stage) are selected as research objects, image data are collected, key points are extracted by YOLOv7-pose network model, and a lightweight convolution method GSConv is introduced, which is able to reduce the complexity of the model and improve the accuracy of the key point detection, the average accuracy of key point detection was improved from 93.7% to 98.7%, an improvement of 5%. Finally, the angular size of the pinch angle was determined based on the coordinates of the key points located by the model, and the R2 of the pinch angle calculation results to the actual manual measurements was the highest of 0.84, the errors for the four periods were 1.543°, 1.716°, 1.213°, and 0.998°. So the method ensures the accuracy of identification while the model size is reduced, and can effectively measure the angle of rice stem and leaf pinch, and the investigated method can better help farmers to understand the growth and development of plants and provide support for the selection of excellent crop traits.
Read full abstract