Redes neurais artificiais para previsão de qualidade de fibra do algodoeiro (Gossypium hirsutum L.)

Abstract

The formation and regulation of cotton fiber quality has always been an important field of research to improve fiber quality, commodities and market value of the crop. Uniformity, length, resistance and micronaire represent the most important factors of cotton fiber quality. Temperature and precipitation in abundance or deficiency are among the most important key factors that affect fiber quality. In the present study, we used different artificial neural network (ANN) architectures to explore the effects of ambient temperature and precipitation on cotton fiber quality, based on 5 cultivars (TMG 44 B2RF, FM 983 GLT, FM 906 GLT, FM 985 GLTP, FM 911 GLTP), in addition to 11,042 HVI (High Volume Instrument) samples from the 2020/21 crop season, in Coromandel, Minas Gerais state. The results show that the ANN model with the 12–12– 1 structure can predict fiber strength with high accuracy (R² = 0.90), the ANN model with the 10–11–1 structure can predict fiber length. fiber with the highest accuracy (R² = 0.91), and the ANN model with the structure of 9–10–1 can predict fiber uniformity with the highest accuracy (R² = 0.95). Predictions from established artificial neural network models can improve the quality of cotton fiber produced on the property by optimizing according to cultivar, in addition to monitored temperature and precipitation.