Inteligência artificial e imagens termográficas infravermelha para avaliação de secagem de sementes

Abstract

Reducing losses and increasing productivity are two factors that need to be overcome. Seed science and technology are essential in boosting the high productive yield of crops. With this, it is sought through the insertion of technology, promising results with practical methods in solving the problems generated in the seed chain. The objective of this work was to evaluate the use of technological innovations through machine learning and infrared thermography in the seed drying process. Data from convective dryers were collected to evaluate the drying performance in machine learning dryers, based on: i) the total number of dryers worked; ii) drying duration time (hours); iii) the percentage of moisture at the entrance and exit of the product (%), and iv) differences in humidity between both (%). Filtered Clusterer, K-means, and Resample filter training models were used to group data based on their similarities. The algorithm and the filters proved efficient in unsupervised classification, identifying and minimizing the fixed system's inter-cluster similarity and defining distinct classes within the data set. To analyze the effects of drying efficiency using a thermal camera with a spectral range of 7.5-13 μm, radiation by three wavelength-dependent fractions, absorbance (a), reflectivity (r), and transmittance (t). The dryer measuring 6.65 x 2.50 x 3.25 m was used, with a 17 t octagon shape. Where it was possible to identify problematic temperature failures during drying, it is concluded that both technologies are promising for evaluating the drying process, facilitating data interpretation, and detecting thermograms.