DeepPalm: Deep Learning Classification Model for Date Palm Varieties

Abstract

Dates being UAE's national fruit, as well as KSA and UAE's national tree hold great importance. Dates are in high demand because of their distinctive flavor, health benefits, and religious significance. Dates come in a variety of flavors, textures, sizes, and shapes. Pakistan produces around 300 different date kinds, of which Dhakki, Aseel, and Begam Jangi are the most common. The choice and quality of this fruit in the local market are mostly determined by human visual perception. The analysis of the quality and selecting the wrong variety concerning demand leads to a loss of product quality and dishonor in the case of export. Intelligent machines have served humanity in these last decades in different aspects of life. Image processing technology has been used extensively in the medical field as well as in agriculture. Identification and classification represent a major challenge for machine learning to achieve almost human recognition levels. This research aims to develop an intelligent method, which would be able to identify and classify date types according to form, size, and color characteristics using machine learning and deep learning. As a first step, the pre-processing technique will be adopted to improve the quality of datasets. The processed images will then be categorized with an appropriate classifier. The proposed system is intended to classify different kinds of dates. To effectively learn and distinguish the types of date fruits with high accuracy, a convolutional neural network (CNN) was constructed and trained from scratch, with nearly 9200 date fruit images which include 1500 images of each of six varieties. The improved model's classification accuracy was 98%. These findings proved the CNN model's ability to recognize types from dates with high accuracy so that it can be used at the industrial level.