Implementation of Semi-Regulated network creation and feature extraction

Abstract

EfficientNet-L2 is a powerful model used for analyzing medical images, especially those of the retina. It improves performance by adjusting three key factors: depth, width, and resolution. Here depth indicates the total number of layers available in the network. Width show the total count of features processed in each layer and resolution gives the clarity of the input images. The adjustment of these key factors helps the model to recognize and process complex details in images more effectively. In this study, we apply EfficientNet-L2 to enhanced retinal images to detect areas affected by diabetic retinopathy. DR is an eye disease caused by diabetes that can cause a loss of vision. We use enhancement techniques such as morphological transformations and Contrast Limited Adaptive Histogram Equalization (CLAHE) to prepare the image-dataset for analysis. These methods helps to improve the visibility of small details in the retina, making it easier to identify affected areas.EfficientNet-L2 is particularly useful because it builds on existing knowledge from pre-trained models and systematically adjusts scaling. It can extract important features from images and can recognize complex patterns giving high accuracy. The model can detect small changes in the retina by training on enhanced images and predict disease progression more effectively. This study combines advanced image processing, feature extraction, and optimized training techniques to create a system that is both accurate and efficient. The results indicate that the model is effective in detecting diabetic retinopathy at an early stage. Thus it helps doctors to provide timely treatment. This approach can also be used for other medical imaging tasks, making EfficientNet-L2 a valuable tool in healthcare. At the end this research shows how using advanced models like EfficientNet-L2 can help improve medical image analysis and lead to better healthcare results.