Performance Evaluation of Deep Learning Architectures for Recognition of Moisture in Dried Coconut Copra

Abstract

Copra is a useful by-product of coconut and is a rich source of edible oil. To make copra, the coconuts are broke opened, and, after the water is drained, the coconut kernels are dried either by exposing to sunlight or by heating in a kiln. However, the traditional sun drying can be used only on sunny days, and it required a minimum of seven to eight days to completely dry the copra. In kiln drying, the coconuts are exposed to flame by placing them on the racks just above the fire. As a result, polycyclic aromatic hydrocarbons (PAH) are deposited on the kernels, which is hazardous to health as PAH is carcinogenic and causes cancer. However, as the coconuts are dried as batches either in the kilns or in the biomass solar dryers, there is every possibility that a good number of copra shells stealthily come out with moisture content not suitable for good food grade quality. Further, these copra shells with moisture content may develop fungus and spread it across the entire batch resulting in huge yield loss. So, there is a need for a low-cost portable system which considers a batch of copra shells to measure the moisture content, classifies them according to the percentage of moisture present in them, and grades them according to the quality based on the moisture content. This paper proposes to develop a computer vision-based Copra Moisture Detection System which implements deep learning architectures to classify the copra according to the moisture content present in the coconut shell.