Search

In this paper, we focus on the segmentation of clouds in All Sky Images using a UNet- based Deep Learning model and the subsequent recognition of the same cloud in different images. This research lays the foundation for the development of solar radiation forecasts with All-Sky Imagers. The implemented model initially extracts relevant features from the input image using convolutions, thereby reducing the resolution. In the subsequent step, the resolution is restored to its original level using transposed convolutions. Contours are then created from all segmented clouds. Using these contours as references, the same cloud is identified in images from different All-Sky Imagers through template and contour matching. We demonstrate that this segmentation approach yields good results on a small test dataset. Additionally, the recognition of clouds in images from different cameras show promising results, with 75 % of clouds being correctly matched. [ABSTRACT FROM AUTHOR]

Buildings are objects of great importance that need to be observed continuously. Satellite and aerial images provide valuable resources nowadays for building footprint extraction. Since these images cover large areas, manually detecting buildings will be a time-consuming task. Recent studies have proven the capability of deep learning algorithms in building footprint extraction automatically. But these algorithms need vast amounts of data for training and they may not perform well under the low-data conditions. Digital surface models provide height information, which helps discriminate buildings from their surrounding objects. However, they may suffer from noises, especially on the edges of buildings, which may result in low boundary resolution. In this research, we aim to address this problem by using edge bands detected by a deep learning model alongside the digital surface models to improve the building footprint extraction when training data are low. Since satellite images have complex backgrounds, using conventional edge detection methods like Canny or Sobel filter will produce a lot of noisy edges, which can deteriorate the model performance. For this purpose, first, we train a U-Net model for building edge detection with the WHU dataset and fine-tune the model with our target training dataset, which contains a low quantity of satellite images. Then, the building edges of the target test images are predicted using this fine-tuned U-Net and concatenated with our RGB-DSM test images to form 5-band RGB-DSM-Edge images. Finally, we train a U-Net with 5-band training images of our target dataset, which contain precise building edges in their fifth band. Then, we use this model for building footprint extraction from 5-band test images, which contain building edges in their fifth band that are predicted by a deep learning model in the first stage. We compared the results of our proposed method with 4-band RGB-DSM and 3-band RGB images. Our method obtained 82.88% in IoU and 90.45% in F1-score metrics, which indicates that, by using edge bands alongside the digital surface models, the performance of the model improved 2.57% and 1.59% in IoU and F1-score metrics, respectively. Also, the predictions made by 5-band images have sharper building boundaries than RGB-DSM images. [ABSTRACT FROM AUTHOR]

Automatic computerized segmentation of fetal anatomical structures from ultrasound images and biometric measurements is still challenging, due to the inherent characteristics of fetal ultrasound images at different semesters of pregnancy. Generally, clinicians manually measure the fetal anatomical structures by fitting either an ellipse (AF-Pocket plane or a line (for the case of femoral plane), and it can take considerable time to take these measurements. The difficulty in automated fetal anatomical structures measurement stems from various uncertain factors, including poor and non-uniform contrast across the border of the fetal structures, the high variability of the ultrasound images, and large intersect variation. It is extremely difficult to extract the edge of the fetal anatomical structures using local patterns alone. In this paper, we propose an automated method for AF-Pocket semantic segmentation using deep learning methods. The proposed method contains four steps: Pre-processing, binary classification, semantic segmentation and AF-Pocket measurement. The results of our proposed architecture were evaluated using a prepared dataset and achieves the best results compared with a variety of state-of-the-art deep learning-based approaches. [ABSTRACT FROM AUTHOR]

Recently, there has been a significant amount of attention towards computer vision algorithms, particularly those that focus on semantic segmentation applications. This is due to the availability of big data to train models, as well as the computational ability of these algorithms. Among the various computer vision applications, the U-Net has gained popularity due to its reliable accuracy, simplicity in construction, and ease of application. Despite the advantages of this network structure, there are still some unclear aspects within the U-Net that have not been significantly covered in literature — to the best of our knowledge —. This study seeks to clarify and explain these ambiguous points and construct different architectures to demonstrate their pros and cons. Finally, a series of experiments were carried out on natural leather samples from MVTec AD to confirm our findings. The outcomes highlight our discoveries and provide a framework for determining the fine-tuning parameters of U-Net. [ABSTRACT FROM AUTHOR]

Road traffic fatalities are a significant concern worldwide, as highlighted by data from the World Health Organization (WHO) and other international organizations. One approach to enhancing road safety is through the assessment of specific characteristics or features that contribute to the overall safety condition of roads. The International Road Safety Assessment Program (iRAP) identifies several attributes that have a direct impact on road safety. Some of these attributes can be collected from satellite imagery. One of first steps in using satellite imagery as source for road attributes collection is road extraction. Quality road extraction can provide a quality base for detection of road attributes. In this paper Random forests, Extreme Gradient Boosting and U-net algorithms were analyzed to get insight into which one is most suitable for road extraction. Analysis was performed on very high-resolution satellite imagery with four spectral bands and spatial resolution of 0.3m. Analysis has shown that U-net outperformed Random forests and XGBoost in each of evaluation measures and it is suggested as best option for road extraction as support of road infrastructure assessment process. [ABSTRACT FROM AUTHOR]

It is important to make car-drivers improve their way of looking for recognizing key objects or areas precisely. This study designs a system following such a motivation that distinguishes several areas in a display with weights of importance. A present proposing function for successful area detection offers drivers an opportunity to compare their gaze with experts. Concrete method for this implementation includes U-Net that is one of major techniques of machine learning combined with grid segmentation. [ABSTRACT FROM AUTHOR]

In order to accurately identify various physiological activities and movements in living bodies, we propose a U-Net-based method to identify all similar micro molecules such as cells and proteins in organisms. We first transform the molecular image to be observed into the feature space using the U-Net convolution network, and then use the target feature to match across the whole image to detect all similar targets of interest in the image. Extensive experimental results show that the proposed method can rapidly detect similar molecules of interest through a simple human-computer interaction and attain a more accurate detection performance than other approaches. [ABSTRACT FROM AUTHOR]