Qualidade de fibra do algodoeiro: Integração entre inteligência artificial e fenotipagem digital

Abstract

The genetic breeding of cotton (Gossypium hirsutum L.) is fundamental for improving the quality and productivity of the world's leading textile fiber, especially in the face of challenges like climate change and increasing demand. The integration of advanced technologies, such as artificial intelligence (AI) and digital phenotyping, is revolutionizing this process by optimizing the selection of desirable traits. This study aimed to demonstrate the potential of two innovative approaches for characterizing cotton fiber quality. First, we explored the use of machine learning models to predict fiber characteristics (such as micronaire, strength, and uniformity) from climatic variables. The Random Forest model showed higher accuracy, with precipitation and minimum temperature in specific months being the most influential variables. We concluded that, despite their potential, including management data could significantly increase the accuracy of the models, making them more robust. Next, we evaluated digital phenotyping based on images to quantify the intensity of fiber coloration. Image analysis allowed for the distinction of genotypes by color saturation. We found that the upper third of the plant produced fibers with lower color intensity. Furthermore, indices derived from image analysis (such as the BI index) showed a positive correlation with important fiber characteristics like uniformity, strength, and length. In summary, this research highlights the value of integrating predictive methods and digital phenotyping tools to accelerate cotton genetic breeding, providing precise and efficient information for decision-making.