Selective Kernel-DenseNet for Effective Areca Plant Disease Classification

Abstract

Accurate classification of Arecanut plant disease is essential for crop damage prevention, ensuring healthy yields and sustainable farming. Timely identification of diseases enables farmers to take prompt action to minimize yield losses and improve overall plantation management. However, plant diseases vary in scales from fine details like spots or small lesions to larger areas of discoloration which often leads to inaccurate performance. To address this issue, this research proposes a Selective Kernel-DenseNet (SK-DenseNet) for an efficient and accurate Arecanut plant disease classification. In traditional DenseNet, the selective kernel is incorporated to enhance feature adaptability by enabling the dynamic adjustment of receptive fields. This enhances the model’s ability to capture both large-scale patterns and fine-grained information, resulting in better feature representation. The dense connections in DenseNet assist in reusing these features effectively which enables the model to handle diseases with symptoms at varying scales. The resizing technique is applied to standardize input dimensions, and label encoding is used to convert categorical into numerical data for effective model processing. When compared with existing methods like VGG19-ViT, the proposed SK-DenseNet obtained a high accuracy of 99.99% and 98.92% on PlantVillage and Arecanut datasets, respectively.