Intelligent Fruit Maturity Assessment Platform Using Convolutional Neural Network: IFMAP

Abstract

As the fight against climate change and the environmental crisis continues, the need for solutions to reduce our carbon footprint increases dramatically. In this paper, we propose a platform to help prevent food waste among Canadian households using an innovative artificial intelligence (AI) system that implements Convolutional Neural Networks (CNN) to analyse pictures of fruits taken with cell phones to determine their current maturity state. Following these results, we have created a native mobile application for the platform which is detailed in the article. We also explain the overall workflow of the platform as well as the different neural networks used and tested for this task. Afterwards, we discuss the use of a wide range of sensors to give more insights about the fruit to our neural networks. We then show a use case of this application on a banana, which was analysed accurately by the neural networks. Finally, we display different results we obtained for the CNNs such as the top performance of different architectures, which can classify the data up to the accuracy of 97.11%. We also demonstrate that a CNN (LeNet) training time is three times faster on a GPU compared to a CPU.