Address:

    Room No.- ECE 1110, Electronics and Communication Engineering Discipline, Dr. Syttendranath Academic Building (Academic Building 1), Khulna University, Khulna 9208.

    Email:

    Contact:

    +8801948820119

    Personal Webpage:
    click here

Classifying Native Fruits using Ensemble Convolutional Neural Network

Abstract: Classification is the process of predicting the class of given data point. Object classification using image is the process of labeling image with correct labels. Classifying objects with high accuracy is one of the most challenging and valuable tasks nowadays because of increasing dependency on machines and computers. Convolutional Neural Networks are the highest performing neural networks in case of classifying objects. In this paper we are exploring a noble approach of ensemble CNN to classify four Bangladeshi fruits. In this study we have used publicly available data and preprocessed them using sample-wise center, sample-wise standard normalization and adaptive histogram technique for feeding four different CNN. We have experimented using different initial learning rate and batch Size to observe performance accordingly. Then we took best three combination of previously mentioned parameters and classify the evaluation set of data independently. We have taken the result and counted a prediction valid only if at least two of them predicted correctly. The proposed method resulted in highest accuracy of 97% and lowest accuracy of 93%. To evaluate the performance, we generated confusion matrix and easily determined our objective. Finally, we have proposed the best strategy which can be used for solving similar problems.

Details
Role Co-Supervisor
Class / Degree Bachelor
Students

Faysal Ishtiaq Rabby [120909]

Start Date 6th June, 2018
End Date 10th December, 2019