Convolutional Neural Network (CNN) architecture for tropical fruit classification

Shaari, Muhammad Haziq (2022) Convolutional Neural Network (CNN) architecture for tropical fruit classification. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Convolutional Neural Network (CNN) architecture for tropical fruit classification.pdf - Submitted Version Download (3MB)

Abstract

Convolutional Neural Network (CNN) is a robust computer vision algorithm from artificial intelligence (AI) that use deep learning networks to analyse visual imagery. The predecessor, machine learning, is also a type of AI but with a lower performance than deep learning. The aim of this project is to design a CNN model architecture layer on par with pre-trained models such as ResNet50, VGG16, MobileNetV2, and DenseNet121 using 15 classes of tropical fruit with total of 15000 images. The model was designed on Google Colab using TensorFlow and Keras library. Next, the network model needs to be trained, test, and optimized by tuning the hyperparameters. Then, all simulation results were compared to the well-known pre-trained models in terms of accuracy and confusion matrix. At the end of this project, the H-CNN able to reach 90.13% accuracy with balance distribution across all of the tropical fruit classification.

Actions (login required)

View Item