A blood orange computer vision sorting system

To have a proper sorting system with a low error rate can be very useful in automatic packaging of products. Furthermore, physical dimensions and shape are important in sorting and sizing of fruits and vegetables. In this paper Iranian orange (blood orange) are considered to present an automatic mass sorting system with low error rate using image processing coupled with an adaptive neurofuzzy inference system (ANFIS). Linear regression analysis was used to compare results and an efficient algorithm was designed and implemented in MatLab. This algorithm is able to measure area, eccentricity, perimeter, length/area, red, green, and blue RGB components, width, contrast, texture, width/area, width/length, roughness and length. In ANFIS model, samples were divided into two sets: 70% for training and 30% for testing. Best ANFIS, linear and nonlinear regression models, yielded values of the coefficient of determination (R2), sum squared error (SSE), and mean squared error (MSE) of 0.989, 21.46, 1.65 (ANFIS), 0.91, 1156.69, 12.05 (linear) and 0.88, 1538.10, 15.86 (nonlinear), respectively. Based on results, ANFIS model showed clearly better capability for mass prediction compared to both linear and nonlinear regression. A prototype for an automatic non-intrusive orange mass sorting system is depicted to conclude.