Improved Convolutional Network With Transfer Learning and Texture Feature Extractor for Plant Disease Detection

Abstract

Global agriculture is seriously threatened by plant diseases, which have an effect on output and food security. For disease care to be effective, prompt detection and precise diagnosis are essential. Traditional methods reliant on the visual inspection are labour-intensive and subjective. Recent technological advances in computer vision and machine learning offer promising solutions. This paper introduces the Transfer Learning-based Plant Disease Detection (TL-PDD) framework, which integrates preprocessing, segmentation, feature extraction and disease prediction stages. Initial preprocessing employs median filtering for data refinement. Segmentation, utilising the Adaptive Pixel Integration in Joint Segmentation (APIJS) approach, isolates disease-affected regions in plant images through a variant of DJS. Feature extraction includes the extraction of critical attributes such as Multi-texton, PHOG and Niblack's Method Assisted in Local Gabor Increasing Pattern (NMA-LGIP). Disease prediction employs a novel Double Convolutional Activation in Convolutional Neural Network-Transfer Learning (DCA-CNN-TL) model, facilitating disease classification and severity assessment based on extracted features. The efficiency and precision of plant disease detection systems can be improved by this framework, supporting efforts to ensure global food security and sustainable agriculture.