The Design of Diabetic Retinopathy Classifier Based on Parameter Optimization SVM

Abstract

Diabetic retinopathy is a kind of disease which can seriously damage eyesight. Early diagnosis and regular treatment can effectively reduce visual deterioration. Artificial judgment of fundus images is time-consuming and easy to misdiagnose. Machine learning is an algorithm which automatically analyzes rules from data and uses rules to predict unknown data. Support Vector Machine (SVM) is one of the most important methods of machine learning. SVM is a classifier with learning ability. It is broadly applied to image recognition and image processing. Based on machine learning, a parametric optimized SVM classifier for diabetic retinopathy is proposed. Firstly, the classifier uses PCA and KPCA method to extract the prominent features of the image without artificial recognizing the features of the image, eliminates the specific feature extraction method, reduces the algorithm complexity, increases the generalization ability of the algorithm, and greatly improves the image processing speed. Secondly, grid search and genetic algorithm are used to optimize the parameters, avoid the problem of slow operation speed and low classification accuracy due to the large amount of data or the unsuitable selection of kernel parameters. Finally, a combinatorial optimization algorithm of KPCA and grid search is created. Meanwhile, the designed experiments verify that this combination optimization algorithm can make the classifier achieve the best classification state. The experimental results show that the classification accuracy of this combinatorial optimization algorithm reaches 98.33%, which can realize the automatic classification of diabetic retinopathy more accurately and rapidly.