| Abstract: | The paper presents an improved method for identification and classification of foreign bodies mixed food grain image samples using a Neural Network Approach. Any matter other than major food grains is considered as a foreign body in this work, such as stones, soil lumps, plant leaves, pieces of stems, weed, other types of grains etc. The amount of foreign bodies decides the purity of the food grains. In manual system human inspectors remove the foreign bodies from the samples and evaluate the grains. In Machine Vision System it is necessary to automatically determine the amount of foreign body present in food grains to help farmers in sowing and also marketing. Here inpainting technique is used to replace the foreign bodies. Different food grains like wheat; groundnut, green gram, jawar and rice are considered in the study. The color and textural features are presented to the neural network for training and later identification of the unknown grain types mixed with foreign bodies. The combination of both color and texture features is employed in the work. The study reveals that the presence of even 10 percent of foreign bodies within food grain image samples reduces its identification and classification accuracies as low as 60%. When the foreign body percentage is greater than 50, it becomes difficult to identify and classify food grain image samples. The identification and recognition performance is improved by inpainting.
|