A two-camera machine vision approach to separating and identifying laboratory sprouted wheat kernels

• A two-camera machine vision is used to classify sprout-damaged wheat.
• Wheat damage was assessed on measured alpha-amylase activities.
• At least 95% of the joined kernels are segmented automatically for classification.
• Neural network with 16 visual properties produced 72.8% classification accuracy.
• Reduction of the system cost and improvement on accuracy are discussed.

A proof-of-concept for a two-camera machine vision (MV) to classify the laboratory sprouted wheat kernels into sound, sprout-damaged, or severely sprout-damaged classes was developed. The marker controlled watershed segmentation technique was tailored to disjoin the clustered kernels in the digital images. Images were captured for both dorsal and ventral sides of the kernels. Segmentation accuracy of at least 95% was achieved depending upon the sample size under some conditions. Sixteen features comprising of colour, texture, and shape and size were extracted from the images of dorsal and ventral sides of the test kernels. The alpha-amylase activities, a key enzyme found in the sprout-damaged wheat were measured analytically for each kernel to categorise it into one of the three classes. A neural network model was developed with the kernels' features as the inputs and the alpha-amylase activity as the output. The MV with the trained neural network classified the test wheat kernels into three classes with an accuracy of 72.8%. Some of the challenges associated with the system are discussed. Recommendations to improve the system accuracy and robustness, and to decrease the system cost are presented.