Copra Dryness Assessment using ResNet50

Copra, the dried coconut kernel from which coconut oil is derived, is typically dried using methods such as sun drying, smoke drying, or kiln drying. Copra quality is primarily determined through visual inspection, assessing factors like moisture content, oil content, free fatty acid (FFA), and meat color. While previous studies have utilized advanced machine learning models for copra dryness assessment, their complex architectures pose challenges for mobile deployment due to high computational requirements. This study aims to address these challenges by using transfer learning with ResNet50 to create a portable, lightweight copra dryness assessment system. The system will be deployed on a Raspberry Pi 4 with a webcam, making it feasible for real-time use in farms. The study will assess three categories of copra dryness—over-dried, optimally dried, and under-dried—based on images collected from local farms. A portable copra dryness assessment system was successfully developed using a Raspberry Pi 4B with a web camera, utilizing transfer learning with ResNet50 to classify copra dryness based on user-captured images. The system achieved a classification accuracy of 91.43%, as determined through a confusion matrix analysis.