Advancing Plant Disease Detection with Hybrid Models: Vision Transformer and CNN-Based Approaches

Early plant disease detection is of prime importance in the current era when agriculture plays a vital role in ensuring food security and sustainable farming. Plant disease detection extends its significance beyond the boundaries of agriculture, with far-reaching implications. It contributes to economic stability, sustainable farming practices, reduced pesticide usage, product quality assurance, global trade compliance, climate resilience, and the advancement of scientific innovations. In the context of Convolutional Neural Networks (CNNs), despite producing astonishing results in computer vision (CV) applications, they still fall behind with their current constraints, such as the limited global content and overfitting issues. These shortcomings can almost be recovered when CNNs are merged with other deep or machine-learning models, including Transformers. Transformers have created a benchmark in the context of finding long-range dependencies in relevance to natural language processing and this success paved the way for their extension to CV tasks. When applied to CV, transformers are simply known as Vision Transformers (ViTs), though they still have their own set of discrepancies when used independently such as issues with spatial hierarchy, computational complexity, object location, and data prerequisites. This chapter focuses on the potential and studies regarding hybrid CNNs and ViTs. So far the development of these models is gaining significance due to their capability of overcoming their inherent inadequacies, thus complementing each other and thereby developing novel opportunities in CV tasks. This chapter also discusses the prospects for hybrid model development, with an emphasis on how they pertain to growing areas such as agriculture, autonomous systems, and healthcare. Many complicated challenges can be tackled through this integration, henceforth changing how the general models understand and interact with the imagery data in a wide range of applications. These models not only prove to be effective for plant disease detection to maintain sustainability and increase productivity in agriculture, but they also offer solutions for progress across other domains.