In this research, digital image processing method (DIPM) is used as an innovative approach to predict precisely the shape of electrical tree (ET) in cross-linked polyethylene (XLPE) power cables in the presence of air voids based on the field calculation using finite-element method (FEM). With the help of DIPM, two case studies are held to detect the accurate parameters of either the first initiated major branch or the tips of the major branches of ET. A hyperbolic needle-to-plane simulation model is proposed to illustrate the ET inception and propagation stages. The non-uniform electric fields thatare accompanied with the electrical treeing phenomenon are calculated using FEM as one of the most effective numerical methods to deal with non-uniform shapes. The predicted shapes of ET initiation and growth are provided in an innovative manner with the implemented hybrid connection between FEM and DIPM for the two proposed case studies. Direction branching approach and deviation angle branching approach are provided in this work to predict the shape and the direction of ET branched voids. The validity of the proposed model is assessed with the help of available previous experimental and simulation data.

This work introduces and evaluates a model for predicting driver behaviour, namely turns or proceeding straight, at traffic light intersections from driver three-dimensional gaze data and traffic light recognition. Based on vehicular data, this work relates the traffic light position, the driver's gaze, head movement, and distance from the centre of the traffic light to build a model of driver behaviour. The model can be used to predict the expected driver manoeuvre 3 to 4 s prior to arrival at the intersection. As part of this study, a framework for driving scene understanding based on driver gaze is presented. The outcomes of this study indicate that this deep learning framework for measuring, accumulating and validating different driving actions may be useful in developing models for predicting driver intent before intersections and perhaps in other key-driving situations. Such models are an essential part of advanced driving assistance systems that help drivers in the execution of manoeuvres.

A novel learning-based end-to-end network for stereo matching, named Multi-path Attention Stereo Matching (MPA-Net), is introduced in this study. Different from existing methods, the multi-path attention aggregation module is designed firstly, named MPA, which is a unified structure using three different parallel layers with a respective attention mechanism to extract the multi-scale informational features. Secondly, the method of cost volume construction, which differs from the traditional stereo matching methods, is extended. And then, the absolute difference between two input features is calculated. Furthermore, a u-shaped structure with 3D attention gate is selected as the encoder-decoder module. Specifically, the module is used to fuse the encoding features to their corresponding decoding features under the supervision of the authors' attention gate with skip-connection, and thus exploit more significant information for matching cost regularisation and disparity prediction. Finally, specific experiments are conducted to evaluate their network on SceneFlow, KITTI2012 and KITTI2015 data sets. The results show that their method achieves a better improvement in disparity maps prediction compared with some existing state-of-the-art methods on KITTI benchmark.

The query image is usually a simple and single object in image retrieval, and the reference images in the database usually have many distractions. The precision of image retrieval can be greatly improved If the target regions in the database image are extracted during retrieval. So this paper proposes a Bow image retrieval method based on SSD target detection. First, the training gallery is manually annotated to record the location and size information. Second, the SSD target detection model is trained with the labeled training gallery to obtain the target object SSD model. Third, the SSD model is used to locate the similar target regions of the reference image and the query graph. Finally, the target region information is mapped into the convolutional features, and these feature vectors are used for image similarity matching. The performance of the proposed method is evaluated on Paris6k, Oxford5k, Paris106k and Oxford105k databases. The experimental results show that the accuracy of image retrieval will be greatly improved by adding optimization methods in the proposed image retrieval framework. The image retrieval accuracy of this method is higher than that of similar methods in recent years.

Currently, the existing literature on multipurpose watermarking employs multiple watermarks for simultaneous achieving of multiple security objectives. This approach can be problematic since a watermark targeting tampers might fail to detect tampering of another watermark inserted for copyright violation. Moreover, most of the current schemes follow a non-blind approach where the original watermark is needed for authentication. A question that naturally arises is whether multipurpose watermarking can be realised with the insertion of a single watermark in a blind way or not. The goal of the authors study is to provide an affirmative answer to this important question. They show how a cryptographic primitive called verifiable threshold secret sharing can be used to come up with a generic construction for blind multipurpose watermarking which inserts a single watermark into the host image for simultaneous achieving of copyright protection, authentication, and tamper localisation. The generic property of the proposed scheme provides flexibility for choosing the embedding/extraction of the watermark based on the desired level of fidelity, robustness and capacity. Experimental results are provided to confirm the superiority of their proposed technique as compared to existing approaches.

Retinal vessel segmentation has important application value in clinical diagnosis. If experts manually segment the retinal vessels, the workload is heavy, and the result is strong subjectively. However, some existing automatic segmentation methods have the problems of incomplete vessel segmentation and low-segmentation accuracy. In order to solve the above problems, this study proposes a retinal vessel segmentation method based on task-driven generative adversarial network (GAN). In the generative model, a U-Net network is used to segment the retinal vessels. In the discriminative model, multi-scale discriminators with different receptive fields are used to guide the generative model to generate more details. On the other hand, in view of the uncontrollable characteristics of the data generated by the traditional GAN, a task-driven model based on perceptual loss is added to traditional GAN for feature matching, which makes the generated image more task-specific. Experimental results show that the accuracy, sensitivity, specificity and area under the receiver operating characteristic curve of the proposed method on data set digital retinal images for vessel extraction are 96.83, 80.66, 98.97 and 0.9830%, respectively.

The existing single image super-resolution methods based on deep learning cannot handle multiple degradations well, and the generated image tends to be blurred and over-smoothed due to poor generalisation ability. In this study, the authors propose a method based on a generative adversarial network (GAN) to deal with multiple degradations. In the generator network, blur kernel and noise level are used as input through dimensionality stretching strategy preprocessing to make full use of prior knowledge. In addition, three discriminators with different scales are used in the discriminator network to pay attention to the reconstruction of image details while focusing on the global consistency of the image. For the problems of vanishing gradient and mode collapse existing in GAN-based methods, a gradient penalty term is added in the loss function. Extensive experiments demonstrate that the proposed method not only can handle multiple degradations to obtain state-of-the-art performance, but also deliver visually credible results in real scenes.

Currently, PCA (principal component analysis) is widely used in many neural networks and has become a crucial part of the convolutional neural network (CNN) feature extraction. However, whether PCA is suitable for this process remains to be elucidated. The authors proposed a new method called balanced principal component (BPC) that generates a balanced local feature and combines with CNN as a layer to cope with the fusion problem. Specifically, BPC layer includes regionalisation module and average compression PCA (AC-PCA) module. First, they used regionalisation module to generate some sub-region that focuses on the local feature in each view. Secondly, the AC-PCA module is a computational process that enlarges the feature matrix by PCA and eventually compacts the matrix to a one-dimensional (1D) vector by AC. Next, all 1D vectors are compacted by AC to obtain a multi-dimensional balance. Finally, they designed this layer with an end-to-end trainable structure to promote the feature extraction task of CNN. They addressed 3D shapes using a projection method that is pre-trained on ImageNet and migration learning on ModelNet dataset. By comparing with the state-of-the-art network, they achieved a significant gain in performance of retrieval and classification tasks.

In order to increase the detection effect of fuzzy vehicles and small size vehicles, an improved multitask cascaded convolutional neural network (IMC-CNN) based on mixed image enhancement is proposed. Firstly, contrast limited adaptive histogram equalisation and multi-scale Retinex are used to enhance images. Mixed image enhancement can effectively solve the problems of image blurring, low contrast and uneven illumination when the imaging environment is not ideal. IMC-CNN includes two stages: object location and object classification. The object location network based on multi-layer feature fusion can locate and extract the object from complex background, and output regions contain only a single vehicle object. The object classification network is a lightweight convolutional neural network with only two convolutional layers, which can effectively reduce information loss and improve the classification accuracy of small objects and fuzzy objects. In addition, online hard example mining algorithm and focal loss function are adopted in network training. These strategies can solve the problem of unbalance between positive and negative samples. To verify the validity of the proposed algorithm, the experiments are performed on SYIT-Fuzzy dataset and COCO-Vehicle dataset. Compared with Faster R-CNN, YOLO v4 and other recent models, the average classification accuracy of the proposed method is significantly increased.

Outdoor images are having several applications including autonomous vehicles, geo-mapping, and surveillance. It is a common phenomenon that the images captured outdoor are prone to noise, which arises due to natural and manmade extreme atmospheric conditions such as haze, fog, and smog. Importantly in autonomous vehicle navigation, it is very important to recover the ground truth image to get the better decision by the system. Estimation of the transmission map and air-light is very crucial in recovering the ground truth image. In this study, the authors proposed a new method to estimate the transmission map based on a mean channel prior (MCP), which represents the depth map to estimate the transmission map. The authors proposed a deep neural network to identify the hazy image for the further dehazing process. In this study, the authors presented, two novel contributions, first an MCP-based image dehazing and second, a deep neural network-based identification of hazy images as a pre-processing block in the proposed end to end system. The proposed deep learning network using the TensorFlow platform provided validation accuracy of 93.4% for hazy image classification. Finally, the proposed MCP-based dehazing network showed better performance in terms of peak-signal-to-noise ratio, structural similarity index, and computational time than that of existing methods.