Computer vision in agriculture, application development using open source tools and systems

Mihály Tóth, Dániel Dér, Szilvia Borbásné Botos, Róbert Szilágyi

Abstract


Nowadays the optimization of agricultural production is a crucial task. The use of computer vision may serve in increasing efficiency as this technology has achieved significant results in many research and practical applications to date. In the following years, we will experience more and more use cases of the technology and its usability in the intensification of agricultural production to meet the demand of the growing population. Computer vision appears as a sub-domain, also in the new and popular concept of Industry 4.0, ensuring an integrated aspect of the technology. Our practical experiment was performed to examine the utility of the currently available open-source toolkits in commuter vision, utilizing OpenCV and Google TensorFlow libraries. In this experiment, the typical processes of computer vision were implemented using various algorithms for each step, including imaging, pre-processing, post-processing and finally, classification. For the experiment, pictures of apples have been used as training data, representing various conditions. The steps including processing, segmentation, and identification of the fruit, were presented. The most commonly used detection algorithms were tested to determine estimated size, shape, and texture properties. Using a convolutional neural network, the identification of the fruit was presented with a recognition accuracy greater than 93%.

Full Text:

PDF


DOI: 10.17700/jai.2019.10.2.545

Journal of Agricultural Informatics