Development of a low cost machine vision system for sorting of tomatoes

Sorting of tomatoes has been an issue faced by producers as well as sellers due to the sheer volumes handled and the delicate nature of the fruit. This paper describes the development of a low cost machine vision system using webcams and image processing algorithms for defect detection and sorting of tomatoes. In the case of agricultural products, good efforts and appropriate techniques are necessary to distinguish between defected and good ones when using machine vision for sorting. Tomatoes having two major defects namely Blossom End Rot (BER) and Cracks could be separated from good tomatoes with calyx. The sorting decision was based on three features extracted by the image processing algorithms. The color features were used for detecting the BER from good tomatoes and shape factor combined with the number of green objects was used for differentiating the calyxes from crack defects. Two methods, rule based and neural network approaches, were developed for decision based sorting. A control system was developed with a belt conveyor to transport the tomatoes and a cylinder pushrod coupled to a solenoid was used to push the defective tomatoes after determining their defect by the algorithms. The color image threshold method with shape factor were found efficient for differentiating good and defective tomatoes. The overall accuracy of defect detection attained by the rule based approach and the neural network method were 84 and 87.5% respectively. The inspection speed of 180 tomatoes min^-1 was achieved by the algorithms and the prototype developed. Comparison of the results obtained by the rule based and neural network approaches are also presented in this paper.