Classification of plant diseases in images using dense-inception architecture with attention modules

Agriculture plays a vital role in providing food to a growing world population. However, plant diseases and pests result in 50% reductions in crop yields, which exacerbates poverty and threatens a sustainable food system. Currently, plant disease detection mainly relies on manual visual inspection and analysis of the field, which is a time-consuming and unreliable method. Modern technology, such as smart agriculture, can leverage machine learning, computer vision, and remote sensing technologies to overcome the limitations of traditional methods. Several machine learning algorithms have been used for plant disease classification and detection. However, these models suffer from limitations such as dependence on complex hand-crafted features for classification, and limited adaptability to new data. Deep learning models, on the other hand, have the ability to automatically learn hierarchical representations from raw data, making them inherently more informative and adaptable to changing data patterns. While existing deep learning approaches for plant disease detection have achieved some degree of success, these models still face challenges in achieving superior performance due to similar visual appearances of different plant diseases, intra-class differences, and variation in scale and resolution. To address these challenges, we proposed a framework that consists of hybrid network and attention module. Our hybrid module combines the benefits of DenseNet and InceptionNet to capture multi-scale features, while the attention module addresses the challenges of inter-class similarity and intra-class variability by isolating the most discriminative feature and disregarding extraneous and insignificant information. We performed experiments on publicly available challenging dataset. From Experimental results, we demonstrate the effectiveness of the proposed framework in achieving high accuracy in plant disease detection.