Classification of Nipa Fruit using Artificial Neural Network

Abstract

The use of computer vision is increasing in agriculture, especially in fruit classification. The different types of fruits make classification difficult. Considering the complex appearance of the images of the fruit, selecting the most suitable data features and tuning hyperparameters is also vital. Hence, image processing is proposed since it offers a fast, efficient, and capable of replacing labor work in the harvesting process. This study was conducted to implement a classification model that uses four types (mature, immature, rejected, and damaged) of nipa fruit using an artificial neural network and to explore the different methods in the classification. There are two methods used in predicting the performance of the model, which are the largest method and the sequential method. These methods were used with four binary neural networks as base learners, which included a three-layer perceptron for each base learner consisting of (Model 1: Mature or Not-Mature, Immature or Not-Immature, Reject or Not-Reject, and Damage or Not-Damage). The largest method accuracy was 78%, the same as for the sequential method (78%). Furthermore, the results show that all the implemented methods outperform the benchmark (Human Classification) by 77%, and experiments also show that other (Multiclass) architectures developed with an accuracy of 86% outperform the remaining classifiers.