Análise e predição da qualidade de grãos de soja por meio de imagens multiespectrais: uma abordagem baseada em deep learning

Abstract

This work explores the importance of assessing the commercial quality of soybeans using a chemometric approach that involves the capture of multispectral images. These images provide a wealth of information surpassing that of conventional images, as they encompass 31 different wavelengths. Consequently, it becomes possible to observe physical and chemical characteristics that would be imperceptible to the naked eye. Additionally, we propose the development of a deep learning model for the automatic and swift classification of beans into predefined categories. The selection of soybeans as the subject of our study is justified by its status as a highly valued legume in global nutrition, coupled with its substantial production, particularly in Brazil. Our database will be constructed with the assistance of a low-cost equipment developed by openScience, offering an opportunity for small and medium-sized enterprises to enhance the efficiency of their grain trading processes.