Identification of maize and wheat seedlings and weeds based on deep learning

Abstract

Introduction: It is well-known that maize and wheat are main food crops in the world. Thus, promoting high quality and abundant maize and wheat crops guarantees the development of the grain industry, which is needed to support world hunger. Weeds seriously affect the growing environment of maize, wheat, and their seedlings, resulting in low crop yields and poor seedling quality. This paper focuses on the identification of maize and wheat seedlings and field weeds using deep learning. Methods: Maize and wheat seedlings and field weeds are the research objects. A weed identification model based on the UNet network model and ViT classification algorithm is proposed. The model uses UNet to segment images. A Python Imaging Library algorithm is used to segment green plant leaves from binary images, to enhance the feature extraction of green plant leaves. The segmented image is used to construct a ViT classification model, which improves the recognition accuracy of maize and wheat seedlings and weeds in the field. Results: This paper uses average accuracy, average recall, and F1 score to evaluate the performance of the model. The accuracy rate (for accurately identifying maize and wheat seedlings and weeds in the field) reaches 99.3%. Compared with Alexnet, VGG16, and MobileNet V3 models, the results show that the recognition effect of the model trained using the method presented in this paper is better than other existing models. Discussion: Thus, this method, which accurately disambiguates maize and wheat seedlings from field weeds can provide effective information support for subsequent field pesticide spraying and mechanical weeding.