研究目的
To compare the recognition level of diabetic retinopathy images based on the combination of Principle Component from 4 features using SVM (Support Vector Machine) method.
研究成果
The model with the highest accuracy and fastest recognition time was the PC1 and PC2 pair, achieving 100% accuracy in training images and the fastest recognition time, indicated by the smallest number of support vectors. The worst performance was observed in the PC3 and PC4 pair. Future research could explore the use of standard deviation or variance for matrix reduction to better represent the data distribution.
研究不足
The recognition accuracy decreased in the testing phase to 93.75%, possibly due to the expansion of the coverage area. The selection of mean values in PCA before matrix reduction might have contributed to this limitation.
1：Experimental Design and Method Selection:
The study used the image of the yellow canal with Gabor filter. Characteristics taken from each image include mean, variance, skewness, and entropy, followed by feature extraction with PCA.
2：Sample Selection and Data Sources:
Database images were obtained from the Eye Centre of Sultan Agung Hospital Semarang, consisting of JPG images with dimensions of 700×605 pixels, 8 bit.
3：List of Experimental Equipment and Materials:
Non-Mydriatic Retinal Camera CR-DGi was used for capturing fundus images.
4：Experimental Procedures and Operational Workflow:
The process included pre-processing, feature vector extraction using PCA, and classification using SVM.
5：Data Analysis Methods:
The accuracy of the classification was evaluated using confusion matrix and the percentage of correct predictions.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
