IEEE Transactions on Pattern Analysis and Machine Intelligence

In the last few years, the deep learning (DL) computing paradigm has been deemed the Gold Standard in the machine learning (ML) community. Moreover, it has gradually become the most widely used computational approach in the field of ML, thus achieving outstanding results on several complex cognitive tasks, matching or even beating those provided by human performance. One of the benefits of DL is the ability to learn massive amounts of data. The DL field has grown fast in the last few years and it has been extensively used to successfully address a wide range of traditional applications. More importantly, DL has outperformed well-known ML techniques in many domains, e.g., cybersecurity, natural language processing, bioinformatics, robotics and control, and medical information processing, among many others. Despite it has been contributed several works reviewing the State-of-the-Art on DL, all of them only tackled one aspect of the DL, which leads to an overall lack of knowledge about it. Therefore, in this contribution, we propose using a more holistic approach in order to provide a more suitable starting point from which to develop a full understanding of DL. Specifically, this review attempts to provide a more comprehensive survey of the most important aspects of DL and including those enhancements recently added to the field. In particular, this paper outlines the importance of DL, presents the types of DL techniques and networks. It then presents convolutional neural networks (CNNs) which the most utilized DL network type and describes the development of CNNs architectures together with their main features, e.g., starting with the AlexNet network and closing with the High-Resolution network (HR.Net). Finally, we further present the challenges and suggested solutions to help researchers understand the existing research gaps. It is followed by a list of the major DL applications. Computational tools including FPGA, GPU, and CPU are summarized along with a description of their influence on DL. The paper ends with the evolution matrix, benchmark datasets, and summary and conclusion.

/pdf/review-of-deep-learning-concepts-cnn-architectures-hpbk62qzc2.pdf

Review of deep learning: concepts, CNN architectures, challenges, applications, future directions

Deep learning is transforming most areas of science and technology, including electron microscopy. This review paper offers a practical perspective aimed at developers with limited familiarity. For context, we review popular applications of deep learning in electron microscopy. Following, we discuss hardware and software needed to get started with deep learning and interface with electron microscopes. We then review neural network components, popular architectures, and their optimization. Finally, we discuss future directions of deep learning in electron microscopy.

Review: Deep Learning in Electron Microscopy

Underwater target detection plays an important role in ocean exploration, to which the improvement of relevant technology is of much practical significance. Although existing target detection algorithms have achieved excellent performance on land, they often fail to achieve satisfactory outcome of detection when in the underwater environment. In this paper, one of the most advanced target detection algorithms, YOLOv5 (You Only Look Once), was first applied in the underwater environment before being improved by combining it with some methods characteristic of the underwater environment. To be specific, the Swin Transformer was treated as the basic backbone network of YOLOv5, which makes the network suitable for those underwater images with blurred targets. It is possible for the network to focus on fusing the relatively important resolution features by improving the method of path aggregation network (PANet) for multi-scale feature fusion. The confidence loss function was improved on the basis of different detection layers, with the network biased to learn high-quality positive anchor boxes and make the network more capable of detecting the target. As suggested by the experimental results, the improved network model is effective in detecting underwater targets, with the mean average precision (mAP) reaching 87.2%, which makes it advantageous over general target detection models and fit for use in the complex underwater environment.

/pdf/underwater-target-detection-algorithm-based-on-improved-nkt9xd60.pdf

Underwater Target Detection Algorithm Based on Improved YOLOv5

Automated driving and vehicle safety systems need object detection. It is important that object detection be accurate overall and robust to weather and environmental conditions and run in real-time. As a consequence of this approach, they require image processing algorithms to inspect the contents of images. This article compares the accuracy of five major image processing algorithms: Region-based Fully Convolutional Network (R-FCN), Mask Region-based Convolutional Neural Networks (Mask R-CNN), Single Shot Multi-Box Detector (SSD), RetinaNet, and You Only Look Once v4 (YOLOv4). In this comparative analysis, we used a large-scale Berkeley Deep Drive (BDD100K) dataset. Their strengths and limitations are analyzed based on parameters such as accuracy (with/without occlusion and truncation), computation time, precision-recall curve. The comparison is given in this article helpful in understanding the pros and cons of standard deep learning-based algorithms while operating under real-time deployment restrictions. We conclude that the YOLOv4 outperforms accurately in detecting difficult road target objects under complex road scenarios and weather conditions in an identical testing environment.

https://www.mdpi.com/2079-9292/10/16/1932/pdf

Road Object Detection: A Comparative Study of Deep Learning-Based Algorithms

With the rapid development of deep learning techniques, deep convolutional neural networks (DCNNs) have become more important for object detection. Compared with traditional handcrafted feature-based methods, the deep learning-based object detection methods can learn both low-level and high-level image features. The image features learned through deep learning techniques are more representative than the handcrafted features. Therefore, this review paper focuses on the object detection algorithms based on deep convolutional neural networks, while the traditional object detection algorithms will be simply introduced as well. Through the review and analysis of deep learning-based object detection techniques in recent years, this work includes the following parts: backbone networks, loss functions and training strategies, classical object detection architectures, complex problems, datasets and evaluation metrics, applications and future development directions. We hope this review paper will be helpful for researchers in the field of object detection.

A review of object detection based on deep learning

Exploiting confident information for weakly supervised prostate segmentation based on image-level labels

The invention provides an indoor scene image migration segmentation method based on a deep weighted full convolution network. The method comprises the steps of for an indoor scene segmentation problem, migrating the full convolutional network model to a segmentation field of the indoor scene, and using a weighted cross entropy loss function during the training process, so that the problem of image class imbalance during the training process is solved, the training is faster and more accurate, and the model is more accurate. Meanwhile, an original data set is increased by generating the image translation and the horizontal reflection, the robustness and the accuracy of the model are improved. According to the method, the indoor scene can be accurately segmented, the visual perception information is provided for the movement and the object grabbing of an intelligent robot, and the method has the higher application value in the field of the indoor scene.

Indoor scene migration segmentation method based on deep weighted full convolutional network

Style transfer is an increasingly popular field that can capture the styles of a particular artwork and use them to synthesize a new image with specific content. Previous NST algorithms have the limitation to transfer styles to correct regions in the output image. Therefore, some regions in the output image have deformed structures of the source image. In this paper, we propose a color preprocessing-based neural style transfer method to overcome the limitation. To reduce impacts caused by color differences between source image and style, we propose three models based on a color iterative distribution transform algorithm (IDT). The first one is named original color-preprocessed (OCp) model, which uses IDT to transform the color probability density function (PDF) of source image into that of style image. The second one is named exposure-corrected original color-preprocessed (EC-OCp) model, which adds an automatic detail-enhanced exposure correction module before OCp model. When source image is underexposed, EC-OCp model can achieve better results than OCp model. The third one is style color-preprocessed (SCp) model. It uses IDT to transform the color PDF of style image into that of source image. The original structures are well protected in the output image. According to experiments, the proposed models are robust to the source images with more conditions. Therefore, they have more usage values than the original method.

Image Artistic Style Transfer Based on Color Distribution Preprocessing

With regard to the problem of image super-resolution (SR), generative adversarial network (GAN) can make generated images have more details and better effect on perceptual quality than other methods. However, GAN-based methods may lose the contour of object in some texture-intensive areas. In order to recover contour better and further enhance perceptual quality, we propose a Multiresolution Mixture Generative Adversarial Network for Image Super-Resolution (MRMGAN), which employs a multiresolution mixture network (MRMNet) for image super-resolution. The MRMNet is able to have multiple resolution feature maps at the same time when training. Meanwhile, we propose a residual fluctuation loss, which aims to reduce the overall fluctuation of residual between SR image and high-resolution (HR) image. We evaluated the proposed method on benchmark datasets. Experimental results show that the proposed MRMGAN can get satisfactory performance.

Yinshu Zhang

Papers

A review of object detection based on deep learning

Exploiting confident information for weakly supervised prostate segmentation based on image-level labels

Indoor scene migration segmentation method based on deep weighted full convolutional network

Image Artistic Style Transfer Based on Color Distribution Preprocessing

Multiresolution Mixture Generative Adversarial Network For Image Super-Resolution