Search

Accurate location information has significant commercial and economic value as they are widely used in intelligent manufacturing, material localization and smart homes. Magnetic sequence-based approaches show great promise mainly due to their pervasiveness and stability. However, existing geomagnetic indoor localization methods are facing the problems of location ambiguity and feature extraction deficiency, which will lead to large localization errors. To address these issues, we propose a coarse-to-fine geomagnetic indoor localization method based on deep learning. First, a multidimensional geomagnetic feature extraction method is presented which can extract magnetic features from spatial and temporal aspects. Then, a hierarchical deep neural network model is devised to extract more accurate geomagnetic information and corresponding location clues for more accurate localization. Finally, localization is achieved through a particle filter combined with IMU localization. To evaluate the performance of the proposed methods, we carried out several experiments at three trial paths with two heterogeneous devices, Vivo X30 and Huawei Mate30. Experimental results demonstrate that the proposed algorithm can achieve more accurate localization performance than the state-of-the-art methods. Meanwhile, the proposed algorithm has low cost and good pervasiveness for different devices. [ABSTRACT FROM AUTHOR]

Deep learning based on convolutional neural network (CNN) has been successfully applied to stereo matching as it can accelerate the training process and improve the matching accuracy. However, the existing stereo matching framework based on CNN often has two problems. The first problem is the generalization ability of training model. Stereo matching frameworks are usually pre-trained on a large synthetic Scene Flow dataset and then fine-tuned on evaluation dataset. However, the evaluation dataset may contain trivial training data or even do not have disparity label for some specified tasks. This adversely affects the generality of the training model. The second problem is the poor matching performance in ill-posed regions. It is difficult to distinguish the ill-posed regions, including weak texture area, repeated texture area, occlusion area, reflection structure, and fine structure, etc. To ameliorate the aforementioned problems, we propose the cost volume enhancement network (CVE-Net) guided by sparse features for stereo matching. CVE-Net use the edge information and saliency information for sparsely sampling the precise disparity labels during training. Furthermore, we enhance the cost volume by leveraging the precise disparity sparse label information to guide the direction of training. The experiment shows that the generalization ability is significantly improved. The domain-transferring problem on the new dataset is significantly alleviated. In addition, introducing the sparse multiple semantic features improves the matching performance in the ill-posed regions. Even without fine-tuning, the matching requirements can be met. These results demonstrate the effectiveness of the CVE-Net. [ABSTRACT FROM AUTHOR]

Representing and analyzing structural differences among graphs help gain insight into the difference related patterns such as dynamic evolutions of graphs. Conventional solutions leverage representation learning techniques to encode structural information, but lack an intuitive way of studying structural semantics of graphs. In this article, we propose a representation-and-analysis scheme for structural differences among graphs. We propose a deep-learning-based embedding technique to encode multiple graphs while preserving semantics of structural differences. We design and implement a web-based visual analytics system to support comparative study of features learned from the embeddings. One distinctive feature of our approach is that it supports semantics-aware construction, quantification, and investigation of latent relations encoded in graphs. We validate the usability and effectiveness of our approach through case studies with three datasets. [ABSTRACT FROM AUTHOR]

Bone age assessment is a pediatric examination that determines the difference between skeletal age and chronological age. The discrepancy between the two ages will often trigger the likelihood of genetic disorders, hormonal complications and abnormalities of maturity in the skeletal system. Recently, although some automated bone age assessment methods by analyzing radiographs have been researched, the available text data from radiological reports are not used. Texts and radiographs are two different modals, the fusion of them can give us much more information for bone age assessment. In this paper, we present a novel multi-modal data fusion-learning network, called RT-FuseNet, for bone age assessment utilizing radiographs and texts. Specifically, we develop a convolutional neural network with spatial pyramid pooling layer and attention mechanism module to ensure the integrity of the image space information and enhance the subtle difference of features among radiographs respectively. In addition, texts are incorporated into the learning model to jointly learn non-linear correlations between various heterogeneous data. To evaluate the proposed approach, two datasets are used and several neural network structures are compared. Experimental results show that the proposed approach performs well. [ABSTRACT FROM AUTHOR]

Medical image segmentation methods based on deep learning show good performance advantages and application prospects, but such methods require experienced medical experts to spend a lot of time labeling data. To address this problem in gland segmentation research, a model named ILECGJL based on incomplete label error correction and group joint learning is proposed, which significantly reduces the dependence on labeled data while ensuring segmentation performance. ILECGJL employs the proposed incomplete label error correction method to capture semantic relationships between labeled and unlabeled images, enhancing the quality of pseudo labels and improving the accuracy of gland segmentation. At the same time, ILECGJL harnesses the proposed group joint learning strategy to enhance the model's robustness on pseudo label noise, further boosting segmentation performance. Experimental results on the CRAG dataset confirm that our proposed label error correction method yields substantial improvements based on initial segmentation predictions. The performance of the ILECGJL model demonstrates significant enhancements compared to fully supervised methods that solely rely on labeled data and outperforms other mainstream semi-supervised approaches. Results and comparisons on the GlaS dataset showcase good generalization ability of the ILECGJL model. Moreover, due to its modular design and customizability, ILECGJL offers high flexibility and can be conveniently integrated into different segmentation models. • Incomplete Label Error Correction and Group Joint Learning model. • A group joint learning strategy for the label error correction. • Gland segmentation on GlaS and CRAG datasets achieves SOTA performance. [ABSTRACT FROM AUTHOR]

We propose a new method for hearing loss classification from magnetic resonance image (MRI), which can automatically detect tissue-specific features in a given MRI. Sensorineural hearing loss (SHNL) is highly prevalent in our society. Early diagnosis and intervention have a profound impact on patient outcomes. A solution to provide early diagnosis is the use of automated diagnostic systems. In this study, we propose a novel Deep-HLNet framework, based on few-shot learning, for the automated classification of SNHL. This research involves magnetic resonance (MRI) images from 60 participants of three balanced categories: left-sided SNHL, right-sided SNHL, and healthy controls. A convolutional neural network was employed for feature extraction from individual categories, while a neural network and a comparison classifier strategy constituted a tri-classifier for SNHL classification. In terms of experiment results and practicability of the algorithm, the classification performance was significantly better than the standard deep learning methods or other conventional methods, with an overall accuracy of 96.62%. [ABSTRACT FROM AUTHOR]

To avoid the exhaustive search over locations and scales, current state-of-the-art object detection systems usually involve a crucial component generating a batch of candidate object proposals from images. In this paper, we present a simple yet effective approach for segmenting object proposals via a deep architecture of recursive neural networks (ReNNs), which hierarchically groups regions for detecting object candidates over scales. Unlike traditional methods that mainly adopt fixed similarity measures for merging regions or finding object proposals, our approach adaptively learns the region merging similarity and the objectness measure during the process of hierarchical region grouping. Specifically, guided by a structured loss, the ReNN model jointly optimizes the cross-region similarity metric with the region merging process as well as the objectness prediction. During inference of the object proposal generation, we introduce randomness into the greedy search to cope with the ambiguity of grouping regions. Extensive experiments on standard benchmarks, e.g., PASCAL VOC and ImageNet, suggest that our approach is capable of producing object proposals with high recall while well preserving the object boundaries and outperforms other existing methods in both accuracy and efficiency. [ABSTRACT FROM AUTHOR]

Highlights • A multi-feature fusion framework is proposed. • A feature representation is developed for HSI using the proposed framework. • State-of-the-art results have been achieved by the proposed method. Abstract Traditional hyperspectral image (HSI) classification methods typically use the spectral features and do not make full use of the spatial or other features of the HSI. To address this problem, this paper proposes a novel HSI classification method based on a multi-feature fusion strategy. The spectral-spatial features are first extracted by spectral-spatial feature learning (SSFL), which is a deep hierarchical architecture. Additionally, the texture features of the local binary pattern (LBP) image are applied and fused with the spectral-spatial features. Then, the kernel extreme learning machine (KELM) is used to classify the hyperspectral images. The results of a number of experiments show that the proposed method effectively improves the classification accuracy of hyperspectral images. [ABSTRACT FROM AUTHOR]

For visually impaired individuals, image captioning is a crucial task that utilizes deep learning models to recognize an image and generate a descriptive sentence, enabling them to understand the content of the image through words. However, the existing image captioning model needs a lot of manual annotation. Fortunately, the emergence of unsupervised methods provides a new approach to image captioning. Our proposed model Fast RF-UIC achieves unsupervised functionality through the designed Pre-trainer. Compared with the existing pre-trained model, the Pre-trainer has a faster and shorter training cycle. The R2-Inception-V4 model is designed as an encoder that fuse the Res2Net structure with Inception-V4 to obtain more image features. Bi-FGRU is designed as the decoder, which the FReLU activation function is used to improve the character representation ability from two-dimensional space. Furthermore, we expanded the corpus used in existing unsupervised image captioning and included additional captions for common objects, effectively enhancing the model's generalization ability. Through experiments, Fast RF-UIC achieved higher scores than existing unsupervised image captioning methods on several text evaluation metrics such as BLUE, ROUGE, and CIDEr. • We propose a fast unsupervised image captioning model RF-UIC. • Our encoder (R2-Inception-V4) and decoder (Bi-FGRU) achieve higher accuracy. • We have expand a dataset of unsupervised image descriptions. [ABSTRACT FROM AUTHOR]

Advances in mathematical models and optimization introduce powerful tools for solving the visual object tracking problem, but this fundamental problem remains unsolved due to various challenges such as illumination variation, deformation and occlusion. Recent progress in correlation filter based models has provided an effective and efficient tracking solution. This type of trackers exploits the property of convolution theorem to significantly speed up their computational processes. However, their scheme of training correlation filters restricts them to use only a few negative examples to train their model, consequently lowers their performance. Therefore, in this paper, we propose to combine the fast calculation advantage of correlation filter with an online discriminative learning model, which can fully exploit the negative examples in the context around the target. The proposed tracker proceeds in a coarse-to-fine scheme: the proposed tracker will first employ a correlation filter to generate a coarse estimation of the location of the target, and then employ a translating model to refine its estimation and calculate the scale variation of the target via a scaling model. Both the translating model and the scaling model are formulated in the framework of our online discriminative model. Besides, we also propose an effective offline representation learning method to generate robust image feature for our correlation filter. Extensive experiments on the online Object Tracking Benchmark against the state-of-the-art methods validate the effectiveness of the proposed tracker. [ABSTRACT FROM AUTHOR]

Fabric image retrieval is beneficial to many applications including clothing searching, online shopping and cloth modeling. Learning pairwise image similarity is of great importance to an image retrieval task. With the resurgence of Convolutional Neural Networks (CNNs), recent works have achieved significant progresses via deep representation learning with metric embedding, which drives similar examples close to each other in a feature space, and dissimilar ones apart from each other. In this paper, we propose a novel embedding method termed focus ranking that can be easily unified into a CNN for jointly learning image representations and metrics in the context of fine-grained fabric image retrieval. Focus ranking aims to rank similar examples higher than all dissimilar ones by penalizing ranking disorders via the minimization of the overall cost attributed to similar samples being ranked below dissimilar ones. At the training stage, training samples are organized into focus ranking units for efficient optimization. We build a large-scale fabric image retrieval dataset (FIRD) with about 25,000 images of 4300 fabrics, and test the proposed model on the FIRD dataset. Experimental results show the superiority of the proposed model over existing metric embedding models. [ABSTRACT FROM AUTHOR]