Aplicação do processamento de imagens e de reconhecimento de padrões com rede neural na estimação da produção de macieiras

Abstract

Technology is an important tool in the global agricultural environment, being responsible for improvements in productivity, with techniques such as precision agriculture, and for the automation of previously manual processes, such as harvesting. Among the various areas of agriculture, fruit production has great global prominence and one of the most common crops in this field is apple. Like all fields of agriculture, the apple is the target of technological investments that seek to improve productivity and the quality of cultivation and life of farmers. Therefore, this work aimed to collect images of apple orchards and, through them, estimate the amount of apples produced. To this end, a photo collection tool was developed and used to obtain these images, which were later unified into different panoramic images. Object detection models were then trained to identify the apples in the photos and, in this way, generate the count. The technique was able to achieve an apple count equivalent to 103.15% of the actual count in the initial test, but was not able to maintain this result for subsequent tests.