Ensuring Food Security and Biodiversity: A Novel Convolutional Neural Network (CNN) for Early Detection of Plant Disease in Precision Agriculture

Abstract

Conventional disease detection methods in agriculture are constrained by the presence of personal opinions and the amount of work required, which hinder broad-scale disease monitoring. This study aims to overcome these difficulties by introducing a biosensor-assisted deep learning system that improves disease identification in precision agriculture. Biosensors, such as hyperspectral or electrochemical sensors, offer an initial means of collecting objective data, which complements subsequent analysis using deep learning techniques. The performance of popular deep learning models (VGG16, MobileNetV2, ResNet50) in classifying diseases across 15 categories is assessed via evaluation on the PlantVillage dataset. In addition, a new Convolutional Neural Network (CNN) structure, which achieves a higher accuracy (99.05%) compared to pre-existing models, is shown. Biosensor data serve as a first screening process, which has the ability to decrease the number of photos that need to undergo deep learning analysis. By utilising this integrated method, the precision and effectiveness of disease identification are enhanced. This framework allows for the early and accurate detection of diseases, which in turn allows for specific therapies and encourages the use of sustainable farming methods. The exceptional precision (99.05%) creates opportunities for practical implementation, perhaps reducing production losses and optimising resource allocation.

Keywords

• plant disease detection
• leaf pathology
• convolutional neural networks
• transfer learning
• deep learning