PLANT DISEASE DETECTION USING CNN WITH XCEPTION ARCHITECTURE

Abstract: This paper constructs a plant disease detection system using Convolutional Neural Networks with the aid of transfer learning on the Xception model. Plant diseases remain one of the critical challenges to agricultural productivity and the detection techniques are often manual inspections by experts which is highly subjective. Our system using the transfer learning and depthwise separable convolutions of the Xception model was able to attain accuracy of more than 99% in identifying 38 different classes of plant disease image lesions from the PlantVillage dataset. The model developed is highly accurate in determining the range of diseases affecting the plants and also distinguishing healthy plants from those which are infected. Based on the experimental results, it can be concluded that the system outperforms traditional approaches that use convolutional neural networks, thus providing a reliable diagnosis tool for farmers and agronomy stakeholders. With the accessible means to swiftly identify a disease, this work serves in showcasing the technology's role in agriculture while aiming to strengthen the loss in crop yields.

Keywords: Plant disease detection, Convolutional Neural Networks, Xception architecture, Transfer learning, Agricultural technology, Image classification