| Abstrakt: |
In DR detection, there are a lot of challenges to be faced in order to provide better performance and accuracy. The problem that still remains in DR detection is selection of image features, and classifiers for appropriate datasets. In order to develop a better detection method, this paper intends to propose an advanced model for detecting DR using fundus images. This detection model accomplishes in four phases include pre-processing, blood vessel segmentation, feature extraction and classification. Initially, contrast limited AHE (CLAHE) and median filtering methods are used for pre-processing. For blood vessel segmentation, fuzzy C-mean (FCM) thresholding works well for making rough clustering of pixels. Further, the local features and morphological transformation-based features are extracted from the segmented blood vessels. Moreover, the deep learning classifier called deep belief network (DBN) classifies the extracted features, which detects whether the image is healthy or affected. As a novelty, the number of hidden neurons in DBN is optimised using modified monarch butterfly optimisation (MBO) termed as distance-based MBO (D-MBO). To the next of the simulation, the performance of the proposed D-MBO-DBN-based DR detection model is compared over the existing models by analysing the most relevant positive, and negative performance measures, and substantiates the overall performance. |
