Guava fruit classification system design with convolutional neural networks

For the rapid and precise advancement of agriculture, artificial intelligence applications are of significant importance. Processes such as disease detection in the agricultural field, identification of soil types, and classification of plants and fruits are currently performed manually. Artificial intelligence enables the automation of these processes, leading to cost reduction and the minimization of human errors. In this study, a system for classifying the species of Guava fruit has been proposed. The proposed system is designed using four pre-trained convolutional neural networks. The convolutional neural networks used are GoogLeNet, Vgg19, ResNet50, and DenseNet201 architectures. The Guava fruit dataset was classified by both k-fold-stratified and an 80:20 split. All experimental studies were evaluated using six different performance metrics. The best result was achieved with the DenseNet201 architecture in the proposed method. The performance results for the DenseNet201 architecture in terms of accuracy, sensitivity, specificity, F1-score, MCC, and kappa are as follows: accuracy - 0.9658, sensitivity - 0.9677, specificity - 0.9954, F1-score - 0.9681, MCC - 0.9640, and Kappa - 0.8268.