Choice of tobacco leaf features based on selected probability of particle swarm algorithm

The image features can be used to grade tobacco leaves automatically. After qualitatively analyzing, dozens of image features affecting tobacco leaf grades are extracted. The influences of these features are different for the tobacco leaf grading. Obviously, removing the less influence factors can improve the speed of tobacco leaf grading without rapidly decreasing the correct grading rate. The speed of tobacco leaf grading is very important because the practicability of tobacco leaf grading refers to not only the correct recognition rate but also the speed. Therefore, a methods based on the feature selected probability of particle swarm is proposed to choose the effective image features used to grade the leaves. The fitness function and the appropriate fitness values θ is determined to choose the seed particles. Then the selected probability of each feature of these seed particles is calculated. Finally the effective features which have the best correct recognition rate for tobacco leaf grading are obtained by determining an appropriate selected probability value. When compared with the MIV, cluster methods, the results of experiments show that the grading speed can be improved by employing the effective features selected by the proposed method under the premise of ensuring a certain recognition rate.