Arecanut Status Detection Using Deep Learning

Abstract

Arecanut, a significant cash crop in many tropical regions, undergoes distinct stages of ripening, posing challenges for timely harvest and market readiness. This research presents a comprehensive framework employing cutting-edge deep learning methodologies, specifically TensorFlow Lite, for accurate and real-time detection of Arecanut status, encompassing ripe, unripe, and dry stages. The integration of OpenCV for image preprocessing and deployment on Raspberry Pi enhances the system's accessibility and usability, enabling on-site detection using the Raspberry Pi camera module. The study begins with a comparative analysis between traditional machine learning techniques, particularly Support Vector Machine (SVM), and deep learning architectures. While SVM exhibited commendable accuracy at 75%, the superior performance and scalability of deep learning models motivated the adoption of TensorFlow Lite. Extensive experimentation, conducted on a diverse dataset of Arecanut images, underscored the efficacy of deep learning in surpassing the accuracy thresholds achieved by conventional approaches. The synergy between TensorFlow Lite and Raspberry Pi leverages the computational capabilities of the latter, facilitating efficient deployment in resource-constrained agricultural settings. This synergy enables real-time inference directly on the device, ensuring data privacy, low latency, and autonomy from cloud connectivity, thus empowering farmers with actionable insights at the point of need, distinguishing between ripe, unripe, and dry stages. The experimental validation of our approach across varied environmental conditions underscores its robustness and adaptability. The deployed system not only offers a portable and cost-effective solution but also contributes to the paradigm shift towards precision agriculture, where datadriven decision-making enhances productivity and sustainability.