Automated Leaf Disease Detection Through Deep Learning

Abstract

Agriculture remains the backbone of many developing nations including Pakistan, yet countless farmers continue to struggle with the early detection of crop diseases due to a lack of reliable and accessible diagnostic tools. This Final Year Project presents a smart, user-friendly mobile application designed to assist in identifying leaf diseases in crops through the power of deep learning. At its core lies a custombuilt convolutional neural network, carefully crafted using advanced components such as bottleneck layers and Squeeze-and-Excitation (SE) blocks to enhance feature extraction and performance. Trained on the widely recognized Plant Village dataset, the model achieved an impressive accuracy of 97.7%, offering farmers a reliable tool to simply capture or upload a photo of a diseased leaf and instantly receive a diagnosis along with recommended treatments. Developed using Android Studio, the app supports both English and Urdu languages, ensuring ease of use for local communities. Additionally, it includes an integrated community feature where users can post updates, share images, and exchange tips with fellow farmers, fostering collaboration, knowledge sharing, and a stronger, more resilient agricultural network.