Strawberries Maturity Level Detection Using Convolutional Neural Network (cnn) and Ensemble Method

Abstract

Harvesting high-quality products at an affordable expense has been the prime incentive for the agriculture industry. Automation and intelligent software technology is playing a pivotal role in achieving both practical and effective solutions. In this study, we developed a robust deep learning-based vision framework to detect and classify strawberries according to their maturity levels. Due to the unavailability of the relevant dataset, we built up a novel dataset comprising 900 strawberry images to evaluate the performance of existing convolutional neural network (CNN) models under complex background conditions. The overall dataset is categorized into three classes: mature, semi-mature, and immature. The existing classifiers evaluated during this study are AlexNet, GoogleNet, SqueezeNet, DenseNet, and VGG-16. To further improve the overall prediction accuracy, two Ensemble methods are proposed based on SqueezeNet, GoogleNet, and VGG-16. Based on the considered performance matrices, SqueezeNet is recommended as the most effective model among all the classifiers and networks for detecting and classifying the maturity levels of strawberries.