Lightweight Deep Learning for Plant Disease Detection: A Review of MobileNet Architectures and TensorFlow.js Deployment

Abstract

The rapid advancement of deep learning has significantly enhanced automated plant disease detection in agriculture. The advancement enables improved crop monitoring and food security management. However, despite high performance in controlled environments, many deep learning models remain difficult to deploy in real agricultural settings due to high computational cost, limited infrastructure, and poor generalization. This paper presents a systematic review of MobileNet-based architectures for plant disease detection, with a focus on their deployment in browser- and mobile-based applications using TensorFlow.js. MobileNet is a lightweight convolutional neural network designed for efficient inference on resource-constrained devices, while TensorFlow.js enables real-time, on-device machine learning without reliance on cloud computing. The review analyzes MobileNet variants (V1, V2, and V3) in terms of accuracy, efficiency, and suitability for deployment. Comparative findings indicate that MobileNetV3 achieves the best balance between performance and computational cost, making it suitable for real-time agricultural applications. However, limitations such as dataset bias, lack of field validation, and efficiency–accuracy trade-offs remain significant challenges. This study provides a deployment-oriented synthesis and identifies key research gaps to support the development of scalable and practical AI-based plant disease detection systems.