Infrared image segmentation of the wind turbine blade is an important link of wind turbine condition monitoring. Nevertheless, there remain plenty of challenges when the traditional semantic segmentation networks are applied directly to the infrared image segmentation of wind turbine blade, including the blurred boundaries caused by similar thermal radiation of background and blade, the irregularity of edge identification caused by uneven heating, and the hub misidentified as blade. To address these issues, a semantic segmentation neural network based on the basic U-Net is improved for infrared image segmentation of the wind turbine blade. The hierarchical-split depthwise separable convolution block is integrated into the constructed network to have a higher segmentation accuracy. Also, the fusion of convolution layer and batch normalization layer is performed in inference to get a better speed–accuracy tradeoff. The experimental results show that the proposed approach outperforms all the comparing methods, which further demonstrates that the trained network can achieve superior runtime performance and accurate segmentation with excellent anti-interference performance.

An Improved U-Net Model for Infrared Image Segmentation of Wind Turbine Blade

Blindness can result from the disease known as diabetic retinopathy, which destroys the retinal blood vessels. Typically, colored fundus injections are used to identify this fatal condition. Taking into account the importance of machine learning algorithms for disease identification, developed a convolutional neural network Visual Geometry Group-16 model to predict diabetic retinopathy illnesses, allowing people to act quickly in terms of treatment. The experimental photographs are divided into four stages: mild, moderate, severe, and proliferative retinopathy. At this stage, micro aneurysms first appear in mild, non-proliferative retinopathy. In moderate non-proliferative retinal disease, a small number of blood vessels that nourish the retina are blocked as the condition worsens. More blood vessels get blocked in severe non-proliferative retinopathy, reducing blood flow to the retina's surface. These regions tell the body to develop new blood vessels so that the area can be nourished. In the advanced stage of proliferative retinal disease, the retina's signals for nutrition cause the development of new blood vessels. Proliferative retinopathy is the name of this condition. These brand-new blood vessels are atypical and delicate. The symptoms of vision loss are not brought on by these blood vessels alone. Since they have fragile and thin walls, blood can leak out and seriously impair eyesight. The results of the studies show that the suggested model may attain an accuracy of 91.06% with a loss in testing of 0.058. This proposed methodology gives a direct route to a deep learning-based diabetic retinopathy prediction system that can be used widely for future purposes.

Employing Deep Learning Algorithm to Identify Diabetic Retinopathy

Diabetic retinopathy is caused by high blood sugar levels, which damage and obstruct the small blood vessels in the retina. The eye creates new blood vessels to make up for the diminished blood flow. These don't work properly and result in blood loss, which impairs eyesight. 1) A vascular hemorrhage might occur if diabetes is not treated and blood sugar levels are not under control. 2) Retinal separation 3) glaucoma-induced blindness. Expertise and technology are generally scarce in areas where the diagnosis of diabetic retinopathy is most important. The majority of research in the field of diabetic retinopathy has been concentrated on disease detection or manually extracting features, but this study hopes to apply deep learning to automatically diagnose the condition in its many stages. This study focuses on the automated identification and categorization of diabetic retinopathy symptoms using a pre-trained convolutional graph neural network (CGNN). A collection of retinal images are used from the eyepacs diabetic retinopathy database., which was categorized into five groups based on the severity of retinal damage. Images from this database were preprocessed and utilized as training and testing data. In addition to being more effective and accurate than manual tests, this method of DR detection also conserves time and money. The transfer learning paradigm used by Alexnet incorporates CGNN technology. In this study, accuracy above 90% is predicted and also plotted the confusion matrix and ROC curve using Matlab.

Detection of Diabetic Retinopathy Image Analysis using Convolution Graph Neural Network

Diabetic Retinopathy ailment is found in human beings causes of high glucose level that leads to vision loss with time. A prior identification of DR symptoms helps to determine the reasons and grading of diabetic retinopathy severity to provide the proper treatment to diabetic patient. Ophthalmologists checkout cotton wool spots, exudates, retinal swellings, micro-aneurysms etc in fundus images. The main work of this paper is to identify these aforementioned lesions, on fundus image dataset having three hundred thirty two images using cornea pannus based augmentation technique. In the medical image processing, Deep learning based classifier model gains better identification of these lesions. VGG-NIN model uses non-linear behavior of fundus images and grades the severity of fundus images into five classes. The performance of proposed model is achieved using parameters accuracy, sensitivity, and specificity. The results prove that proposed diabetic retinopathy grading system acquire 91.25% accuracy, 92.56% sensitivity and 94.87% specificity using 10-cross fold validation method than KNN, CNN, SVM classifier model.

Augmentation Based Diabetic Retinopathy System Using Fundus Images

Various Pre-processing and segmentation techniques for detecting DR in human eyes are discussed in this article. Gray scaling, noise filtering, sharpening, segmentation, and feature extraction and classification are among the Pre- processing approaches. The training phase and the testing phase are the two phases. Each of the Pre-processing strategies for detecting micro-aneurysms and hard exudates is represented by a different icon in a single Graphical User Interface (GUI) window built using Java Swing. The produced image can be examined comfortably at an earlier stage, minimising the amount of time spent examining it. Finally, the suggested technique compares time and accuracy using the following algorithms: CNN, ResNet, MobileNet, and AlexNet, which reduces the training time cost. The GUI displays the comparison of algorithms in the form of bar charts and graphs. Our proposed model achieves approximately 95% accuracy for two class classification and over 85% accuracy for five class classification.

Graphical User Interface for Classification of Diabetic Retinopathy using Image Processing

Diabetic retinopathy (DR) is one of the major complications caused by diabetes and can lead to severe vision loss or even complete blindness if not diagnosed and treated in a timely manner. In this paper, a new feature map global channel attention mechanism (GCA) is proposed to solve the problem of the early detection of DR. In the GCA module, an adaptive one-dimensional convolution kernel size algorithm based on the dimension of the feature map is proposed and a deep convolutional neural network model for DR color medical image severity diagnosis named GCA-EfficientNet (GENet) is designed. The training process uses transfer learning techniques with a cosine annealing learning rate adjustment strategy. The image regions of interest of GENet are visualized using a heat map. The final accuracy, precision, sensitivity and specificity of the DR dataset of the Kaggle competition reached 0.956, 0.956, 0.956, and 0.989, respectively. A large number of experiment results show that GENet based on the GCA attention mechanism can more effectively extract lesion features and classify the severity of DR.

https://ieeexplore.ieee.org/ielx7/6287639/6514899/09664550.pdf

Classification of Diabetic Retinopathy Severity Based on GCA Attention Mechanism

Semantic matching plays a crucial technical supporting role in question answering systems. For semantic matching, it is mainly based on neural networks to solve the sentence representation and interaction in semantic matching. The Siamease network structure is a commonly used structure in semantic matching. In view of the problem of information exchange and semantic extraction not saving points in the independent training of the network structure using shared parameters for the input two sequences. Therefore, in this paper, a multi-head attention mechanism is added to the Siamease model based on bidirectional LSTM to solve the problem of insufficient semantic extraction, and on this basis, it is proposed to add fine-grained matching information in the form of an interactive multi-head attention mechanism to solve the interaction problem. Experimental results show that the performance of the model is further improved compared to previous deep learning models.

Yulin Liao

Papers

Classification of Diabetic Retinopathy Severity Based on GCA Attention Mechanism

Semantic Matching Research based on Interactive Multi-head Attention Mechanism