Comparative Performance Evaluation of Size Metrics and Classifiers in Computer Vision based Automatic Mango Grading

Any horticulture product is expected to meet the set quality standards in order to improve its market value. The mango is one of the most delicious tropical fruit from India. Exports from India account for 0.11 per cent of the total domestic production as against about 4 per cent in case of other exporting countries like Mexico, Philippines and Venezuela. Case study of mango fruit from Maharashtra, India is being considered here for size grading purpose. Images of mango are captured using CCD camera and size analysis is carried out using the MATLAB package. The paper discusses various size estimation metrics which are used as feature vectors for two classifiers namely Feed Forward Neural network (FFNN) and Support Vector Machines. The performance of these classifiers in grading mangoes according to CODEX size standards is presented. Experimental results show that Statistical method give an average size grading efficiency of 97% irrespective of classifiers for mango size grading.