Abstract: The images of intact, mildewed and damaged sunflower seed samples were acquired and preprocessed to extract 3 color features’value of G, B and I, and extract 5 texture feature values which include gray uniformity, gradient uniformity, moment of inertia, consistency and entropy. Then BP neural network and decision tree algorithm were conducted to discriminate above three kinds of sunflower seed samples based on extracted color features and texture features. Results showed that all eight image features were adopted by BP neural network, and the BP model gave the recognition rate by 98.58% for intact, mildewed and damaged sunflower seed samples respectively; while only two features were inputted into decision tree model, G and B, and the model gave the recognition rate by 99.25%. Compared with BP neural network model, the decision tree model shows more concise structure and more efficient performance. Result told that intact, mildew and damaged sunflower seeds can be discriminated computer vision technology.

Key words: sunflower seeds, computer vision, neural network, decision tree