Published Paper Details:

This paper presents the development and evaluation of a mobile application designed for real-time plant disease detection using a convolutional neural network (CNN) model deployed with TensorFlow Lite and implemented through the Flutter framework. The EcoGate Android application is an innovative mobile solution developed using Java/XML and integrated with Firebase Realtime Database to promote sustainable gardening and eco-friendly practices. The app combines multiple intelligent modules into a single platform: an E-commerce marketplace for gardening tools and eco-products, a leaf disease detection system that leverages machine learning for early plant health monitoring, an AI-powered chatbot offering real-time gardening tips and problem-solving guidance, and a video learning module that provides curated gardening tutorials. Users can purchase eco-products, diagnose plant issues by uploading leaf images, receive instant AI-based recommendations, and access educational resources to improve their gardening skills. By integrating real-time data storage and AI-driven features, EcoGate enhances user engagement, fosters sustainable environmental practices, and bridges the gap between technology and eco-conscious living. The importance of early disease detection in plants cannot be overstated, as it plays a crucial role in preventing the spread of diseases and ensuring optimal crop yield. Traditional methods of plant disease detection involve manual inspection by experts, which is time-consuming, subjective, and not scalable for large agricultural operations. Other existing solutions, such as cloud- based machine learning models, require continuous internet access, leading to latency issues and dependency on network availability. These limitations highlight the need for a more efficient and accessible solution. In this context, the proposed mobile application stands out by offering a real-time, offline capability that is both efficient and user-friendly. The application utilizes the camera package in Flutter to access the device's camera and continuously capture frames. These frames are then processed using a TensorFlow Lite model, which has been optimized for mobile devices. The model was trained on a comprehensive dataset consisting of various plant diseases, enabling it to accurately classify and identify disease symptoms from the captured images. The methodology section of this paper details the entire devel- opment process, including dataset preparation, model training, and conversion to TensorFlow Lite. The dataset comprises labeled images of healthy and diseased plants, covering a wide range of common plant diseases such as leaf blight, rust, and powdery mildew. Data augmentation techniques were employed to increase the diversity and size of the dataset, thereby enhancing the model's robustness. The CNN model architecture was chosen for its effectiveness in image classification tasks, and it was trained using TensorFlow with parameters optimized for high accuracy and generalization. Post-training, the model was converted to TensorFlow Lite format, involving quantization techniques to reduce the model size while maintaining performance, thus ensuring smooth and efficient inference on mobile devices. The mobile application development phase leveraged Flutter for its cross-platform capabilities and expressive UI components. The user interface was designed to be intuitive and accessible, with key screens including a home screen, a live scanning interface, and a result display. The home screen provides users with information and instructions, while the scanning interface displays the live camera feed along with real-time detection results. Upon detection of a disease, the application displays detailed information about the disease, including possible treat- ments and preventive measures. Performance evaluation of the application was conducted to assess its accuracy, latency, and user experience. The model achieved an accuracy of 95%, with a precision of 93%, recall of 94%, and an F1-score of 93% on the test dataset. Real-time performance metrics indicated that the application processes frames at a rate of 15 frames per second, with a detection latency of approximately 200 milliseconds. User feedback from preliminary testing highlighted high satisfaction with the app's speed and accuracy, emphasizing its practical utility in real-world agricultural scenarios. The discussion section of the paper analyzes the results, com- paring the proposed solution with existing methods. The proposed mobile application outperforms traditional manual inspection and cloud-based solutions in terms of speed, accessibility, and user convenience. These limitations suggest directions for future research, including improving the model's robustness, expanding the dataset to cover more plant diseases, and integrating addi- tional features such as disease treatment recommendations and a user-friendly interface for educational purposes. In conclusion, this research demonstrates the feasibility and effectiveness of using a mobile application for real-time plant disease detection. By integrating a TensorFlow Lite model with the Flutter framework, the application provides a practical and accessible tool for farmers and agricultural professionals, aiding in early disease detection and contributing to enhanced agricultural productivity. Future work will focus on refining the model and application, aiming to further support the agricultural community in combating plant diseases and ensuring sustainable crop production.