Identification of split and whole cashew nuts based on machine vision

Abstract. Cashew nuts are kidney/concave shaped tree nut which is popular all over the world. During cashew processing, it undergoes various thermal and mechanical stresses and results in the splitting of nuts. Ideally, the market value of whole cashew nuts is higher than the splits (half of whole). Grading cashews into these two major grades (whole, split) is manually done by visual inspection. Machine vision based grading provides an alternative to the manual method; however, the direct imaging may create misclassification as wholes and splits look alike at a certain arrangement. The proposed algorithm in the study is the images of ‘shadows‘ cast by the cashew nuts under point illumination source vary among the wholes and splits that can be used for identification. The algorithm consists of two levels of identification; (i) texture profile based identification to identify split cashews facing up (split-up); (ii) shadow dimension based identification to separate whole cashews and split cashews facing down (split-down). The algorithm was developed in ImageJ, an open source Java based image processing platform. Out of several features extracted (minimum, maximum, average, standard deviation, variance, length of the curve, and area under the curve) and tested from texture profile, length of the curve was found suitable to identify split-up cashews with an accuracy of 100%. New parameters for identifying whole and split-down cashews were developed such as ratio of shadow area to cashew area (S:C), and ratio of shadow area to total area (S:T). Out of these, S:T was found more suitable to identify whole and split-down cashews. The overall accuracy of the algorithm was 97.3% (2 incorrect out of 75 objects). The promising results suggest that the developed algorithm can be coupled with a suitable hardware system to perform accurate separation of the whole and split cashews.