DETERMINING THE STAGES OF STRAWBERRY MATURITY USING CONVOLUTIONAL NEURAL NETWORK MODELS

Abstract

Throughout the history of agriculture, crop cultivation and harvesting technologies have been constantly improving, easing the work in the field and increasing the efficiency of fruit harvesting. Today, modern technologies, including artificial intelligence, are also penetrating the agro-industrial sector. Its use is becoming increasingly widespread every year. This technology allows agricultural producers to obtain large amounts of data in real-time, analyze them, and make decisions regarding fertilizer application, pesticide use, irrigation, and determining the ripeness of fruits or plants. An equally important aspect of artificial intelligence solutions is the tracking of carbon footprints, which gives an advantage when entering European markets. The combination of the latest developments in unmanned vehicles and robots allows for an increase in productivity and production volumes in agriculture. An important part of modern agro-industry is computer vision — one of the fields of artificial intelligence, focused on creating intelligent systems capable of processing and analyzing visual information in a way similar to the human sensory system. Neural networks are often used in this technology for both image recognition and classification. This article presents a comparative analysis of various convolutional neural network models for classifying the ripeness stage of strawberries, namely MobileNetV2, MobileNetV3Small, and EfficientNetB0. The network was evaluated based on parameters such as training and validation accuracy, training and validation losses, and training time. The best results were found with the MobileNetV3Small network. The results and methodology of this research can be useful for both scientists and entrepreneurs working in the agro-industrial sector and implementing artificial intelligence in the production process.