Real-time determination of flowering period for field wheat based on improved YOLOv5s model

Abstract

The flowering period is one of the important indexes of wheat breeding. The early or late flowering affects the final yield and character stability of wheat. In order to solve the problem that it is difficult to accurately and quickly detect the flowering period of a large number of wheat breeding materials, a determination method of flowering period for field wheat based on the improved You Only Look Once (YOLO) v5s model was proposed. Firstly, a feature fusion (FF) method combing RGB images and corresponding comprehensive color features was proposed to highlight more texture features and reduce the distortion caused by light on the extracted feature images. Second, the YOLOv5s model was selected as a base version of the improved model and the convolutional block attention model (CBAM) was adopted into the feature fusion layer of YOLOV5s model. Florets and spikelets were given greater weight along the channel and spatial dimensions to further refine their effective feature information. At the same time, an integrated Transformer small-target detection head (TSDH) was added to solve the high miss rate of small targets in wheat population images. The accurate and rapid detection of florets and spikelets was realized, and the flowering period was determined according to the proportion of florets and spikelets. The experimental results showed that the average computing time of the proposed method was 11.5ms, and the average recognition accuracy of florets and spikelets was 88.9% and 96.8%, respectively. The average difference between the estimated flowering rate and the actual flowering rate was within 5%, and the determination accuracy of the flowering period reached 100%, which met the basic requirements of the flowering period determination of wheat population in the field.