Advancing Coconut Tree Disease Detection Using Integrated CNN And Hybrid Vision Transformer

In the agrarian societies of tropical regions, coconut plantation plays an essential role in ensuring economic stability and agricultural sustainability. One of the toughest challenges in the production of coconut trees is to detect diseases at an earlier stage and with efficiency. Advancements in Deep learning techniques have opened new avenues for efficient plant disease detection in recent years. This paper introduces a methodical strategy using a Hybrid Vision Transformer model, combining Convolutional Neural Networks (CNN) and the Vision Transformer architecture. By combining these two architectures we can leverage strengths of both architectures. A substantial amount of data is required to train any pre-existing model, highly sensitive of class imbalance and also, they are prone to overfitting. The proposed model utilizes EfficientNetB3 for feature extraction and a transformer encoder for sequence modeling. The source of the dataset is the Mendeley dataset repository and consists of 5,798 images. The are classified into five disease categories. The proposed method outperformed other pretrained and traditional models, demonstrating superior performance across all metrics and achieving better results specifically for coconut tree disease detection. The proposed model performed over 98.36% accuracy in all aspects. Overall, this study has exhibited its effectiveness in diagnosing the diseases in Coconut trees.