Advancements in precision agriculture: Edge Impulse-based convolutional neural networks for robust apple disease diagnosis

Abstract

Apple cultivation is essential to the global economy, but it is vulnerable to diseases that can have severe impacts on crop yields and cause economic losses. Accurate and timely detection of these diseases is crucial for effective prevention and management. This study addresses the limitations of traditional machine learning methods, such as manual feature extraction and high computational costs, by employing convolutional neural networks (CNNs) and transfer learning using the MobileNetV2 architecture for the detection and categorization of four common apple ailments: Apple Scab, Black Rot, Cedar Apple Rust, and Healthy apples. The model produced excellent accuracy of 96.7% on the testing dataset while working with an enriched dataset of 12,000 photos in various dimensions. The model's performance on the validation set was 89.7% with a loss of 0.60. The transfer learning model demonstrates superior on-device performance, with faster inferencing time and lower RAM usage, making it suitable for real-world applications in apple orchards. In conclusion, this study successfully demonstrates the potential of using convolutional neural networks and transfer learning with the MobileNetV2 architecture for apple disease detection. The model's high accuracy and performance make it a promising tool for assisting apple growers in monitoring and managing apple diseases, ultimately leading to more efficient and sustainable apple production.