Identification of varieties of wheat seeds based on multispectral imaging combined with improved YOLOv5

The identification of seed variety is important in wheat production because the growth and yield are highly related with its variety. Traditional identification methods for wheat seed varieties were suffered with time consuming and contamination. This study proposed a method for convenient identification of wheat seed varieties by using improved YOLOv5 combined with multispectral images. Three optimal spectral bands images including 405 nm, 570 nm and 890 nm were selected to fuse new image from all 19 bands using Genetic algorithm and confusion matrix. The YOLOv5s model for wheat seed variety identification was improved by adding a convolutional block attention module (CBAM) and the identification model was developed with the fusion images. The identification performance of proposed method achieved an accuracy of 99.38 % in testing set, which was better than traditional VOLOv5 with RGB images or with the multispectral images. Meanwhile, the evaluation indexes of the model such as P/%, R/%, F1/% and mAP/% were all higher than 90 %, which showed that the method was suitable for identification of wheat seeds variety rapidly and non-destructively.