Automated Cashew Nut Grading using Deep Learning Techniques

Abstract

Grading cashew nuts is an integral process and one of the factors that greatly impacts the quality, price and value of cashew products, especially for export. Traditionally, in Sri Lanka, grading entails the manual inspection of cashew nuts using a subjective system that also takes considerable time and provides inconsistent grading. In this study, we presented deep learning systems to cater automated grading systems that are adapted to the local context. A classification system for cashew nuts was proposed, based on four varied categories: diseased, whole, half-split and broken. A new dataset of cashew nut photos from Northern province of Sri Lanka was created and labeled for the first time in the region. A unique image dataset of cashew kernels was created to enable localized model training and evaluation. Deep learning models, including VGG19 for classification and ResNet50 for detailed analysis were implemented to enhance grading accuracy across categories such as whole, half-split, broken and diseased kernels. Next, YOLOv8 nano models were applied for multi-object detection, ensuring precise identification of cashew nuts within images. The performance of the system was evaluated, yielding a mean Average Precision across IoU thresholds from 0.5 to 0.95 of 0.949. The overall mean Intersection over Union (IoU) was 0.9639 and the classification accuracy reached 0.9721, demonstrating the robustness and reliability of the system in automated cashew nut grading. The system was implemented on a Raspberry Pi 4B (4GB) device, with a connected camera for real-time detection and classification, displaying results directly on the screen. A camera mounted above the conveyor belt captures moving cashews, and live video is processed at 30 FPS into individual frames. The YOLOv8-Nano model classifies nuts in real time, with instant grading results displayed on the monitor. The proposed system significantly improves grading consistency, reduces human error and offers a scalable, practical solution for integration into local cashew processing industries.