engleski

RBG camera as a non-destructive tool for nutrient deficiency detection

In view of the increasingly evident climatic changes that affect nature, natural processes and, consequently, all human activities on Earth, which, of course, includes agricultural production as food production factory in openfield, arises the need to apply new techniques and technologies in a modern fruit production. Since we can no longer fully rely on established production processes with established production procedures/measures, it is necessary to control plantations more frequently to identify problems in production related to proper growth and development of the plant and fruit, which mainly concerns proper plant nutrition, as well as plant protection. All this requires additional human involvement and work, and thus additional time. In modern apple growing, efforts are being made to use innovative technologies that optimize production but also reduce human work. One of these innovative technologies can be the use of RGB digital cameras [a camera equipped with a standard complementary metal-oxide semiconductor (CMOS) sensor used to take color photos of objects] as a tool for collecting data in machine learning process. The usage of RGB cameras in machine learning could provide fruit growers to detect and prevent potential damage caused by a deficiency of a particular nutrient in a timely manner. Also, it would allow them to respond in time by taking agrotechnical measures that could mitigate the harmful effects of a potential problem, reduce production costs, and thus increase profits. So, this study aims to investigate the use of the RBG camera as a non- destructive tool to detect nutrient deficiencies for building a model based on machine learning. For this purpose, 5 productive orchards were selected in which 200 images per day were taken within 5-time intervals in one day. The images were captured manually and later processed in an annotation program (LabelImg). On each individual image, all leaves were annotated and classified as either 'healthy leaf' or 'leaf deficient in nitrogen, phosphorus, potassium, calcium, magnesium, iron, zinc, or manganese'. From September 2021 to June 2022, more than 3200 images were collected and almost 7500 annotations were made regarding the nutritional apple leaf status. The most numerous class was healthy leaf with about 6000 annotations. Based on a large number of annotations, a model capable of distinguishing healthy leaves from nutrient deficient leaves will be created. Final goal will be developing software/application and constructing a robot- rover with 4 RGB cameras built-in, that would be able to drive autonomously through the orchard and providing real-time information to the fruit growers on the condition of the orchard. Such continuous monitoring with RGB cameras through the orchard could be a key factor in planning and implementing a greater number of agrotechnical and pomotechnical activities, timely application of fertilizers, rational and more effective use of human work, and ultimately better fruit quality. The application of such new technologies would significantly improve fruit production.

apple, innovative technologies, machine learning, mineral, robot-rover

nije evidentirano