Peanut maturity classification by features extracted from selected hyperspectral components

Effective evaluation of the maturity of seed is critical information to ensure the quality of the crops and provide better economic returns. The traditional assessment method requires manual exocarp removal for color evaluation. This assessment is limited by the observer’s color assessment skill and requires significant time for exocarp removal without damaging the seed. Standard RGB imaging has been proven insignificant in distinguishing peanut maturity. Therefore, Hyperspectral images with a relatively complex linear unmixing-based optimization have been used for the peanut maturity classification with many wavelength channels. To make it affordable to the farmers, cost needs to be minimized as much as we can. Therefore, this paper deals with reducing the number of wavelengths in the hyperspectral camera by feature selection techniques, and input memory storage for classification with a simpler random forest model while maintaining the same efficiency as the linear unmixing model.