"Groundnut stage wise leaf spot disease dataset"

"Plant disease detection plays a critical role in ensuring food security and improving crop productivity, particularly in the agricultural sector. Leaf spot disease significantly affects groundnut, a major oilseed crop, resulting in reduced yield and economic loss. In this context, it focuses on the early detection of groundnut leaf spot using Deep Learning (DL) techniques. In this study, a novel Multi-Scale Lightweight Convolutional Block Attention Module (MS-LiteCBAM) MobileNetV2-based model is proposed for accurate classification of the disease at early stages, enabling timely intervention and control measures. The study compares six pre-trained Convolutional Neural Network (CNN) models. The model uses 8050 realtime images to identify four different stages (early leaf spot, moderate leaf spot, severe leaf spot, healthy leaves) groundnut leaf spot. Among these, the proposed MS-LiteCBAM MobileNetV2 model achieved the highest classification training accuracy and validation accuracy at 99.80% and 99.75%, respectively. Furthermore, the model is deployed in a real-time mobile application to facilitate in field level disease detection. The application not only suggests possible treatments and preventive methods based on the classified disease but also stores the predicted results in real-time using Firebase cloud storage. This enables continuous monitoring, supports data-driven decision-making, and provides valuable assistance to farmers for timely action. Overall, the system enhances groundnut disease diagnosis while promoting precision agriculture and sustainable farming practices."

植物病害检测对于保障粮食安全、提升作物生产力至关重要，尤其在农业领域。叶斑病会显著危害作为主要油料作物的花生（groundnut），导致产量下降与经济损失。在此背景下，本研究聚焦于利用深度学习（Deep Learning, DL）技术实现花生叶斑病的早期检测。本研究提出一种基于多尺度轻量级卷积块注意力模块（Multi-Scale Lightweight Convolutional Block Attention Module, MS-LiteCBAM）的MobileNetV2模型，用于精准分类早期病害，以便及时采取干预与防控措施。本研究对比了6种预训练卷积神经网络（Convolutional Neural Network, CNN）模型。研究采用8050张实时图像，用于识别花生叶斑病的4个不同阶段：早期叶斑、中度叶斑、重度叶斑以及健康叶片。其中，所提出的MS-LiteCBAM MobileNetV2模型取得了最高的分类训练准确率与验证准确率，分别达99.80%与99.75%。此外，该模型已部署至实时移动应用中，以支持田间级病害检测。该应用不仅可基于分类后的病害结果提供相应的治疗与预防方案，还可通过Firebase云存储实时存储预测结果。这一设计能够实现持续监测，支撑数据驱动的决策制定，并为农民及时采取行动提供宝贵的辅助。总体而言，本系统可提升花生病害诊断能力，同时助力精准农业与可持续耕作实践的推广。