Accurate & Efficient Plant Disease Detection using Transfer Learning with Edge Impulse

Abstract

Transfer learning is a powerful machine learning technique that allows optimization of pre-trained models for related tasks on small datasets. In this research paper, we explore the application of Edge Impulse & transfer learning for plant diseases and aim to detect on edge devices more effectively at low cost. We collected and preprocessed many plant images and used this data to fine-tune a neural network model pre-trained by Edge Impulse. We measure the performance of the fine-tuned model on the validation set and optimize for efficiency using Edge Impulse's model optimization feature. Our focus lies in deploying optimized models to edge devices and closely monitoring their performance in real-world environments. Our results show that adaptive learning using Edge Impulse enables us to achieve higher accuracy and efficiency in plant disease detection compared to training from scratch, which is promising for the cultivation of plant diseases. This research contributes to the evolution of learning and edge computing by providing insights and guidance to develop accurate and efficient learning models for diagnosing plant diseases on edge devices.