ABSTRACT :
This paper applies deep learning techniques to the retinal blood vessels segmentations based on spectral fundus images. It presents a network and training strategy that relies on the data augmentation to use the available annotated samples more efficiently. Thus, the shape, size, and arteriovenous crossing types can be used to get the evidence about the numerous eye diseases. In addition, we apply deep learning based on U-Net convolutional network for real patients’ fundus images. As a result of this, we achieve high performance and its results are much better than the manual way of a skilled ophthalmologist.
EXISTING SYSTEM :
Convolutional networks (ConvNets) are a special type of neural network that are especially well adapted to computer vision applications because of their ability to hierarchically abstract representations with local operations. There are two key design ideas driving the success of convolutional architectures in computer vision. First, ConvNets take advantage of the 2D structure of images and the fact that pixels within a neighbourhood which are usually highly correlated. Further, ConvNet architectures rely on feature sharing and each channel (or output feature map) is thereby generated from convolution with the same filter at all locations. This important characteristic of ConvNets leads to an architecture that relies on far fewer parameters compared to standard Neural Networks. Second, ConvNets also introduce a pooling step that provides a degree of translation invariance making the architecture less affected by small variations in position. Notably, pooling also allows the network to gradually see larger portions of the input. The increase in receptive field size (coupled with a decrease in the input’s resolution) allows the network to represent more abstract characteristics of the input as the network’s depth increase. For example, for the task of object recognition, it is advocated that ConvNets layers start by focusing on edges to parts of the object to finally cover the entire object at higher layers in the hierarchy.
DISADVANTAGES OF EXISTING SYSTEM :
1) Less accuracy
2)low Efficiency
PROPOSED SYSTEM :
The architecture of CNN used for the blood vessel segmentation of the fundus images is presented in Figure 5. It was derived from the U-Net network presented in Figure 4. The U-Net exhibits the encoder-decoder architecture where the decoder gradually recovers it. As a result, it produces a pixel-wise probability map instead of classifying an input image as a whole. The U-Net in opposition to other CNN architectures does not require a huge amount of training samples and can be effectively trained with only a few images. This was also in the case of the dataset considered in this study.
Compared to the original architecture, some important modifications were introduced in the CNN used in this work. First, the network was downscaled. Particularly, the depth of the network was reduced by removing two (out of five) levels of pooling/up sampling operations with the corresponding convolution. Additionally, the number of feature vectors at each level was halved. As a result, the number of filters varies from 32 at the input to 128 in the lowest resolution. The downscaling was performed since shallower architecture allowed to obtain equivalent results as the original U-Net, but the training became easier and its time was significantly reduced. The final number of layers and their configuration were selected via experimentation and were balanced between the training time and the accuracy of network. Additionally, dropout layers were introduced between the convolutional layers to improve the training performance.
ADVANTAGES OF PROPOSED SYSTEM :
1) High accuracy
2)High efficiency
SYSTEM REQUIREMENTS
SOFTWARE REQUIREMENTS:
• Programming Language : Python
• Font End Technologies : TKInter/Web(HTML,CSS,JS)
• IDE : Jupyter/Spyder/VS Code
• Operating System : Windows 08/10
HARDWARE REQUIREMENTS:
Processor : Core I3
RAM Capacity : 2 GB
Hard Disk : 250 GB
Monitor : 15″ Color
Mouse : 2 or 3 Button Mouse
Key Board : Windows 08/10
