Banana Disease Recognition Using Deep Learning Based Method Combined With Attention Mechanism

In numerous Asian countries, bananas are a key component of international trade and a significant source of foreign exchange earnings. However, the banana agriculture sector faces a substantial challenge due to disease affecting banana leaves. These diseases significantly affect the sector by reducing both the quality and quantity of the produce. To address this problem, this study proposes a convolutional neural network (CNN) model, enhanced with a spatial attention mechanism as an augmented module for identifying diseases in banana leaves. We compiled a dataset of 5,583 images, from real-life instances and the Mendeley database, divided into 13 classes representing different banana leaf diseases. During preprocessing, a deep convolutional generative adversarial network (DCGAN) was utilized to augment the image samples, thereby addressing the problem of uneven data distribution. In our comparative analysis, we assessed several well-known pre-trained models employing transfer learning and fine-tuning techniques, alongside a vision transformer model VIT32. The experimental results were promising: the proposed model achieved an accuracy of 96.37% indicating its efficacy in classifying various types of banana leaf diseases on this multiclass dataset.