A tomato disease identification method based on leaf image automatic labeling algorithm and improved YOLOv5 model

Tomato is one of the most important vegetables in the world. Timely and accurate identification of tomato disease is a critical way to ensure the quality and yield of tomato production. The convolutional neural network is a crucial means of disease identification. However, this method requires manual annotation of a large amount of image data, which wastes the human cost of scientific research.To simplify the process of disease image labeling and improve the accuracy of tomato disease recognition and the balance of various disease recognition effects, a BC-YOLOv5 tomato disease recognition method is proposed to identify healthy growth and nine types of diseased tomato leaves. In the present study, the YOLOv5 model is improved by designing an automatic tomato leaf image labeling algorithm, using the weighted bi-directional feature pyramid network to change the Neck structure, adding the convolution block attention module, and changing the input channel of the detection layer. Experiments show that the BC-YOLOv5 method has an excellent image annotation effect on tomato leaves, with a pass rate exceeding 95%. Furthermore, compared with existing models, the performance indices of BC-YOLOv5 to identify tomato diseases are the best.BC-YOLOv5 realizes the automatic labeling of tomato leaf images before the start of training. This method not only identifies nine common tomato diseases, but also improve the accuracy of disease identification and have a more balanced identification effect on various diseases. It provides a reliable method for the identification of tomato disease. © 2023 Society of Chemical Industry.