Detecting Tomato Disease Types and Degrees Using Multi-Branch and Destruction Learning

Tomato diseases have a crucial impact on tomato yield and quality, and their detection and control can improve agricultural production efficiency and promote the healthy development of agricultural production. In this study, diseased tomato leaves were used as the research object and fine-grained feature learning of diseases was performed using a deep-learning method to achieve tomato disease degree detection. The proposed model uses ResNet-50 as the backbone network, and the object positioning, destruction and reassembly, and attention area division modules are built on it. Without adding parameters or further boundary annotation, the object positioning module could reliably predict the location of lesions. By erasing the contextual semantic data, the proposed destruction and reassembly module learns more granular features. The attention area division module is capable of localizing the lesion positions, thereby enhancing the recognition accuracy. The results of the study indicate that the proposed model performed better than state-of-the-art methods in identifying the extent of tomato diseases, and achieved an accuracy of 95.03%. The accuracy in identifying the types of tomato diseases is 98.25%. Concludingly, the proposed model is competitive with other state-of-the-art approaches for identifying the severity of tomato diseases and provides a novel model for identifying the degrees of plant diseases.