Light back scattering imaging is an advanced technology applicable as a non-destructive technique for monitoring quality of horticultural products. Because of novelty of this technique, developed algorithms for processing this type of images are in preliminary stage. The present study investigates the feasibility of texture-based analysis and coefficients from space-domain analysis to develop better models for predicting mechanical properties (fruit flesh firmness or elastic modulus) of horticultural products. Images of apple, plum, tomato, and mushroom were acquired using a back scattering imaging setup capturing 660 nm. After segmenting the back scattering regions of images by variable thresholding technique, they were subjected to texture analyses and space domain techniques in order to extract a number of features. Adaptive neuro-fuzzy inference system models were developed for firmness or elasticity prediction using individual types of feature sets and their combinations as input for prediction model applicable in real time applications. Results showed that fusion of the selected feature sets of image texture analysis and space domain techniques provide an effective means for improving the performance of back scattering imaging systems in predicting mechanical properties of horticultural products. The maximum value of correlation coefficient in the prediction stage was achieved as 0.887, 0.790, 0.919, and 0.896 for apple, plum, tomato, and mushroom products, respectively.
