Dates Fruit Classification Using Convolution Neural Networks

Date is a fruit rich in dietary fiber with global commercial value due to its nutritional and medicinal value in curing and preventing various diseases. The grading and sorting process of date fruits is still a complex task in the industry requiring a great deal of human expertise. The breakthrough of deep learning in computer vision has opened up new path for various industrial applications. Therefore, the objective of this paper is to employ convolutional deep learning network and develop an improved automatic sorting system for classifying the images of three main varieties of date fruits, namely, Ekhlas, Nbute Sultan, and Shayshi. To this end, a convolutional neural network (CNN) was designed and trained from scratch to efficiently learn and discern the different varieties of date fruits with high accuracy. The generalization ability of the developed CNN architecture was improved with a combination of convolutional and max-pooling layers. Added to, it was trained from scratch with nearly 3165 date fruit images which include 1055 images of each three varieties. A comparative analysis with different gradient descent optimization methods on the developed CNN architecture is presented. The obtained results testified the efficacy and generalization ability of the developed CNN architecture to discriminate date cultivars with higher accuracy. Therein, the integration of the developed CNN architecture in a suitable affordable framework, will make it possible to reduce the inconsistency and lag of human labor by raising the accuracy and speed of the process.