Performance Optimization of Convolutional Neural Networks (CNN) in Multilabel Fruit Classification

Embedded systems with computer vision via a deep learning approach are becoming increasingly common in a variety of fields, including agriculture, where they can be adapted and used in supermarkets, the food industry and cold stores. By using deep learning such as Convolutional Neural Networks (CNNs) in these sectors, we can improve product quality, reduce waste, ensure regulatory compliance and optimize processes, resulting in significant economic and environmental benefits. However, like any technology, they are not without their challenges. Assessing and improving performance in these systems requires an approach combining model optimization, the use of more efficient specialized hardware and intelligent management of computing resources. In this context, the final objective of our project is to develop and evaluate the performance of an embedded computer vision system via a deep learning-based approach: Application to the sorting and grading of agricultural products. To achieve the final objective, in this article we will first try to present and define the architecture of a type of CNN model, and we will build a fruit classification model based on visual appearance using the 'Fruits-360 dataset-Kaggel '. The deep learning model will be based on Tensorflow and Keras sequential APIs and will aim to efficiently classify different fruit types and predict fruit type.