Fruits Freshness Classification System using Computer Vision and AI on Jetson Nano and Edge Impulse

Abstract

The classification of fruit freshness is commercially important in the food industry. Due to a lack of automation systems in this industry, this process is often done by human labour, resulting in inefficiency and inaccuracy in fruit freshness assessment, which can lead to food waste. As a result, this project will focus on developing a deep learning model that is reliable and accurate at identifying the freshness of fruits and categorizing them as fresh or rotten using two different datasets, as well as evaluating the model's performance in real-time on the Jetson Nano board to set a benchmark against existing models. The model was created using MobileNetV2, a well-known image classification architecture based on Convolution Neural Network (CNN). Data augmentation and pre-processing were already used in both datasets, resulting in 12196 and 13588 images from the Kaggle dataset and the Mendeley dataset, respectively. The model's hyper-parameter was tuned and analysed to classify 6 and 16 fresh and rotten classes from both datasets, respectively. In classifying 6 and 16 classes of various fresh and rotten fruits, the developed model achieved remarkable accuracy of 98.92% and 99.11%, respectively.