Adaptive Neuro-Fuzzy Regression of Coconut Inflorescence Texture Phenotypes Through Spectral Analysis Using Computer Vision

The coconut inflorescence is one of the sources of various food products coconut palm trees (Cocos nucifera L.) have to offer. In line with the advancement of the production of coconut inflorescence by-products and derivatives, the characterization of their texture features through spectral analysis was given importance. It is vital to determine the texture information of the coconut inflorescence in order to identify its maturity, which is a significant factor in harvesting a larger amount of good-quality sap. Using the MATLAB programming language, this study developed five regression-based adaptive neuro-fuzzy inference systems (ANFIS) models that would predict five coconut inflorescence texture phenotypes through spectral feature analysis. A total of 100 images from the three inflorescence maturity level categories were utilized for image processing. With this, spectral and texture features were intelligently extracted. Twelve segment-based spectral features were then used as the input to a feature selection procedure, while five texture parameters were utilized as dependent variables. After running Principal Component Analysis and Neighborhood Component Analysis, the RGB color model was selected to be utilized as an independent variable in the regression system. In conclusion, five ANFIS were able to predict corresponding textures accurately. The model for predicting contrast was the most accurate, while the model for predicting entropy texture was the least accurate. However, system improvement is also recommended by having a more categorically balanced number of images in the dataset.