Deep Level Sets for Salient Object Detection

Deep (machine) learning in recent years has significantly increased the predictive modeling strengths for applications in the areas of computer vision, speech recognition, and natural language processing. Being heavily data-driven, conventional general-purpose deep learning methods mainly rely on training data to learn task-specific data representations. In this thesis, we focus on leveraging novel learning algorithms in deep learning to tackle visual recognition challenges at the three key levels -- namely pixel, object, and image levels. Throughout the thesis, we demonstrate how each and every of our proposed methods can address a specific visual recognition problem. Firstly, we apply deep learning to improve the saliency detection performance of deep networks. A coarse saliency map of the input image's salient object is initially predicted by a convolutional-deconvolutional network (or equivalently a fully convolutional network). Following that, the coarse saliency map is sequentially refined by a separate network called Recurrent Attentional Convolutional-Deconvolutional Network (RACDNN) that shares its network weights across its sequential steps. At every sequential (temporal) step, the same RACDNN is applied to a newly selected/attended image sub-region and it performs saliency map refinement for that particular sub-region. Despite having many learnable network parameters and no access to additional training data, RADCNN can be trained well and it outperforms the "single-step" (coarse saliency detection) baseline and state-of-the-art saliency detection methods by large margins, on several saliency detection benchmark datasets. Secondly, we introduce auto-encoder weight transfer networks (AE-WTN+), a deep model for scaling up object detection. Conventional WTN transfers object class knowledge from the classification weights of a pretrained large-scale image classification network, to the classification weights of an object detection network. However, they suffer from under-fitting and are not able to preserve the rich object class information of novel classes. In this work, we propose a new WTN method -- AE-WTN+ which makes use of normalization techniques to ease network optimization. Inputs and intermediate features are normalized to improve trainability for reduced under-fitting. More importantly, a novel auto-encoding (reconstruction) loss is proposed to encourage information preservation of all classes in the weight predictions of AE-WTN+. This enables the predicted weights of AE-WTN+ to maintain the class discriminability and neighbourhood relationships contained in the pre-trained large-scale classification weights. Experiments on large-scale detection datasets validate the effectiveness of AE-WTN+ on both seen and novel classes. Thirdly, we improve the efficiency of cross-layer network connections for very deep convolutional network in the context of image classification. This gives rise to a new convolutional network architecture called Deluge Networks (DelugeNets) which pass the output (feature activations) of any preceding composite network layer as direct inputs to all succeeding composite layers. For the incoming feature activations of preceding composite layers, the succeeding composite layers first apply rather lightweight cross-layer depthwise convolutional weights on them for feature aggregation. The heavy lifting is carried out by the last convolutional weight layer in each preceding composite layer that performs cross-channel convolutions. The proposed technique enables unimpeded information flow among many layers in very deep DelugeNets with great efficiency. Experiments on small-scale and large-scale datasets show that DelugeNets perform competitively to state-of-the-art deep convolutional networks at lower computational and parameter costs. Fourthly, we propose Stochastic Downsampling (SDPoint), a method to overcome a major limitation suffered by conventional deep convolutional networks -- their computational costs during inference are fixed and cannot be changed. SDPoint downsamples the feature maps at a randomly selected point (layer index) in the network hierarchy during training. The many downsampling configurations known as SDPoint instances each entail a unique downsampling point and downsampling ratio. The same convolutional network weights are shared across the different SDPoint instances that downsample feature maps differently. During inference, the computational cost of a network trained with SDPoint can be instantaneously adjusted to fit a given computational budget, by manually selecting an appropriate SDPoint instance. Additionally, sharing weights between different feature map scales provides significant regularization benefits, making the networks trained with SDPoint less sensitive to scales and more competent. In image classification experiments, SDPoint demonstrates significant cost-accuracy and computational advantages, over independently-trained (non-weight-shared) baselines with various inference costs.

Deep learning has been widely used within learning algorithms for robotics. One disadvantage of deep networks is that these networks are black-box representations. Therefore, the learned approximations ignore the existing knowledge of physics or robotics. Especially for learning dynamics models, these black-box models are not desirable as the underlying principles are well understood and the standard deep networks can learn dynamics that violate these principles. To learn dynamics models with deep networks that guarantee physically plausible dynamics, we introduce physics-inspired deep networks that combine first principles from physics with deep learning. We incorporate Lagrangian mechanics within the model learning such that all approximated models adhere to the laws of physics and conserve energy. Deep Lagrangian Networks (DeLaN) parametrize the system energy using two networks. The parameters are obtained by minimizing the squared residual of the Euler–Lagrange differential equation. Therefore, the resulting model does not require specific knowledge of the individual system, is interpretable, and can be used as a forward, inverse, and energy model. Previously these properties were only obtained when using system identification techniques that require knowledge of the kinematic structure. We apply DeLaN to learning dynamics models and apply these models to control simulated and physical rigid body systems. The results show that the proposed approach obtains dynamics models that can be applied to physical systems for real-time control. Compared to standard deep networks, the physics-inspired models learn better models and capture the underlying structure of the dynamics.

Deep learning is a recently developed feature representation technique for data with complicated structures, which has great potential for soft sensing of industrial processes. However, most deep networks mainly focus on hierarchical feature learning for the raw observed input data. For soft sensor applications, it is important to reduce irrelevant information and extract quality-relevant features from the raw input data for quality prediction. To deal with this problem, a novel deep learning network is proposed for quality-relevant feature representation in this article, which is based on stacked quality-driven autoencoder (SQAE). First, a quality-driven autoencoder (QAE) is designed by exploiting the quality data to guide feature extraction with the constraint that the potential features should largely reconstruct the input layer data and the quality data at the output layer. In this way, quality-relevant features can be captured by QAE. Then, by stacking multiple QAEs to construct the deep SQAE network, SQAE can gradually reduce irrelevant features and learn hierarchical quality-relevant features. Finally, the high-level quality-relevant features can be directly applied for soft sensing of the quality variables. The effectiveness and flexibility of the proposed deep learning model are validated on an industrial debutanizer column process.

Deep learning (also known as deep structured learning, hierarchical learning, or deep machine learning) is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using a deep graph with multiple processing layers, composed of multiple linear and nonlinear transformations. Deep learning has been characterized as a class of machine learning algorithms with the following characteristics [257]: • They use a cascade of many layers of nonlinear processing units for feature extraction and transformation. Each successive layer uses the output from the previous layer as input. The algorithms may be supervised or unsupervised and applications include pattern analysis (unsupervised) and classification (supervised). • They are based on the (unsupervised) learning of multiple levels of features or representations of the data. Higher-level features are derived from lower-level features to form a hierarchical representation. • They are part of the broader machine learning field of learning representations of data. • They learn multiple levels of representations that correspond to different levels of abstraction; the levels form a hierarchy of concepts. These definitions have in common: multiple layers of nonlinear processing units and the supervised or unsupervised learning of feature representations in each layer, with the layers forming a hierarchy from low-level to high-level features. Various deep learning architectures such as deep neural networks, convolutional deep neural networks, deep belief networks (DBN), and recurrent neural networks have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition, and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. In this chapter we start in Section 7.1 with an introduction, giving a brief history of this field, the relevant literature, and its applications. Then we study some basic concepts of deep learning such as convolutional neural networks, recurrent neural networks, backpropagation algorithm, restricted Boltzmann machines, and deep learning networks in Section 7.2. Then we illustrate three examples for Apache Spark implementation for mobile big data (MBD), user moving pattern extraction, and combination with nonparametric Bayesian learning, respectively, in Sections 7.3 through 7.5. Finally, we have summary in Section 7.6.

Short-term power load prediction is not only a critical part of power system dispatching but also an essential task for power marketing, grid planning, and other management departments; deep learning is an artificial intelligence method that has gained extensive attention in recent years; this paper selected three models with the most typical signs of deep learning recurrent neural networks and studied their model performance for short-term power load predictions, including power load data preprocessing, feature selection of prediction models, determination of model parameters, and comparison of prediction characteristics between shallow and deep networks, to explore and prove the applicability of deep learning neural networks in short-term power load predictions, ultimately improving the accuracy of short-term load predictions.KeywordsPower loadDeep learningPrediction

BackgroundTo determine if mammographic features from deep learning networks can be applied in breast cancer to identify groups at interval invasive cancer risk due to masking beyond using traditional breast density measures.MethodsFull-field digital screening mammograms acquired in our clinics between 2006 and 2015 were reviewed. Transfer learning of a deep learning network with weights initialized from ImageNet was performed to classify mammograms that were followed by an invasive interval or screen-detected cancer within 12 months of the mammogram. Hyperparameter optimization was performed and the network was visualized through saliency maps. Prediction loss and accuracy were calculated using this deep learning network. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were generated with the outcome of interval cancer using the deep learning network and compared to predictions from conditional logistic regression with errors quantified through contingency tables.ResultsPre-cancer mammograms of 182 interval and 173 screen-detected cancers were split into training/test cases at an 80/20 ratio. Using Breast Imaging-Reporting and Data System (BI-RADS) density alone, the ability to correctly classify interval cancers was moderate (AUC = 0.65). The optimized deep learning model achieved an AUC of 0.82. Contingency table analysis showed the network was correctly classifying 75.2% of the mammograms and that incorrect classifications were slightly more common for the interval cancer mammograms. Saliency maps of each cancer case found that local information could highly drive classification of cases more than global image information.ConclusionsPre-cancerous mammograms contain imaging information beyond breast density that can be identified with deep learning networks to predict the probability of breast cancer detection.

Though there are many computational models proposed for saliency detection, few of them take object boundary information into account. This paper presents a hierarchical saliency detection model incorporating probabilistic object boundaries, which is based on the observation that salient objects are generally surrounded by explicit boundaries and show contrast with their surroundings. We perform adaptive thresholding operation on ultrametric contour map, which leads to hierarchical image segmentations, and compute the saliency map for each layer based on the proposed robust center bias, border bias, color dissimilarity and spatial coherence measures. After a linear weighted combination of multi-layer saliency maps, and Bayesian enhancement procedure, the final saliency map is obtained. Extensive experimental results on three challenging benchmark datasets demonstrate that the proposed model outperforms eight state-of-the-art saliency detection models.

MFENet: Multitype fusion and enhancement network for detecting salient objects in RGB-T images

Deep neural networks have made significant achievements in representation learning of traditionally man-made features, especially in terms of complex objects. Over the decades, this learning process has attracted thousands of researchers and has been widely used in the speech, visual, and text recognition fields. One deep network multi-layer extreme learning machine (ML-ELM) achieves a good performance in representation learning while inheriting the advantages of faster learning and the approximating capability of the extreme learning machine (ELM). However, as with most deep networks, the ML-ELM’s algorithmic performance largely depends on the probability distribution of the training data. In this paper, we propose an improved ML-ELM made via using the local significant regions at the input end to enhance the contributions of these regions according to the idea of the selective attention mechanism. To avoid involving and exploring the complex principle of the attention system and to focus on the clarification of our local regional enhancement idea, the paper only selects two typical attention regions. One is the geometric central region, which is normally the important region to attract human attention due to the focal attention mechanism. The other is the task-driven interest region, with facial recognition as an example. The comprehensive experiments are done on the three public datasets of MNIST, NORB, and ORL. The comparison experiment results demonstrate that our proposed region-enhanced ML-ELM (RE-ML-ELM) achieves performance increases in important feature learning by utilizing the apriori knowledge of attention and has a higher recognition rate than that of the normal ML-ELM and the basic ELM. Moreover, it benefits from the non-iterative parameter training method of other ELMs, and our proposed algorithm outperforms most state-of-the-art deep networks such as deep belief network(DBN), in the aspects of training efficiency. Furthermore, because of the deep structure with fewer hidden nodes at each layer, our proposed RE-ML-ELM achieves a comparable training efficiency to that of the ML-ELM but has a higher training speed with the basic ELM, which is normally the width single network that has more hidden nodes to obtain the similar recognition accuracy with the deep networks. Based on our idea of combining the apriori knowledge of the human selective attention system with the data learning, our proposed region-enhanced ML-ELM increases the image classification performance. We believe that the idea of intentionally combining psychological knowledge with the most algorithms based on data-driven learning has the potential to improve their cognitive computing ability.

Recently, data-driven algorithms such as deep neural networks have attracted a lot of attention and have become a popular area of research and development. This interest is driven by several factors, such as recent advances in processing power (cheap and powerful hardware), the availability of large datasets (big data), and several small but important algorithmic advances (e.g., convolutional layers). Nowadays, deep neural networks are the state-of-the-art for several computer vision tasks, such as the ones that require high-level understanding of image semantics, e.g., image classification, object segmentation, saliency detection, and also in low level image processing tasks, such as image denoising, inpainting, and super-resolution, among others. Deep learning can handle large volumes of video data by making use of its powerful non-linear mapping and extracting high-level features with very deep networks. Incorporating deep learning into the IoVT can provide radical innovations in video sensing, coding, enhancing, understanding, and evaluation areas making it easier to handle the enormous growth in video data compared to traditional methods.

Salient object detection (SOD) has been attracting a lot of interest, and recently many computational models have been developed. Here, the authors formulate a SOD model, in which saliency map is computed as a combination of the colour, its distribution‐based saliency and orientation saliency. Similar to traditional SODs, the proposed method is based on super‐pixel segmentation and super‐pixel utilises both colour and its distribution‐based saliency to generate a coarse saliency map. However, distinct from traditional SODs, authors further use orientation contrast to optimise the coarse saliency map to obtain an improved saliency map. Authors’ contributions are twofold. First, the authors combine colour uniqueness and its distribution with local orientation information (LOI) used in Itti's model to effectively improve profiles of salient regions. Second, a reciprocal function is defined to substitute the Gabor function used in LOI, and the authors have proved that the substitution could detect relatively homogeneous and uniform regions at the boundary of salient object, whereas it is what the traditional models lack. Authors’ approach significantly outperforms state‐of‐the‐art methods on four benchmark datasets, while the authors demonstrate that the proposed method runs as fast as most existing algorithms.

Nowadays, ship detection in sea-sky background is not only useful in maritime visual surveillance, but also helpful in maritime search and rescue. Since ships are salient objects in infrared images with sea-sky background, we present a novel and effective algorithm based on saliency for ship detection in this situation. Our algorithm adopts global saliency, local saliency and background prior to generate saliency maps. Ships are finally segmented in saliency maps. Our algorithm is compared with four classic salient object detection algorithms. And experimental results show our algorithm outperforms the other four algorithms in qualitative and quantitative terms. Keyword list: ship detection ; salient object detection ; infrared image ; sea - sky background . 1. INTRODUCTION Since ships are important vehicles for sea transportation, ship detection in sea-sky background is extensively used in maritime visual surveillance and reconnaissance, avoidance of ship collision accidents, maritime search and rescue, etc. Infrared imaging systems can almost work in any environment and all-weather conditions with its visual capacity in darkness and frog, which makes it a very suitable tool in this application. In recent years, ship detection in infrared images with sea-sky background attracts increasing attention. Many methods are proposed in this field, including image segmentation methods [1-2] and morphological methods [3]. However, due to complex sea-sky environment and large variation of ship sizes, these methods cannot solve the issue completely. And since ships are often salient objects in infrared images with sea-sky background, some researchers have recently tried salient object detection ideas used in color images [4-6]. And the paper is focused on using salient object detection methods to solve the issue. Salient objects are objects which stand out in the image and attract most people ¶VLPPHGLDWHDWWHQWLRQ7KHVWDQGDUGway to detect salient ob MHFWVLVILUVWO\XVLQJVDOLHQF\GHWHFWLRQPHWKRGVWRJHQHUDWHVDOLHQF\PDSVLQZKLFKDSL[HO¶VJUH\value is proportional to its degree of saliency and then segmenting salient objects in saliency maps. Many salient detection methods can be divided into three categories: local contrast methods, global rarity methods and other methods. In local contrast methods, regions which have high contrast with local neighbouring regions are regarded as salient regions. Itti et al. [7] decompose an image into several feature maps . The saliency in each feature map is measured by local center-surround difference in several scales and saliency of different feature maps are integrated to a final saliency

Automatic extraction of salient regions is beneficial for various computer vision applications, such as image segmentation and object recognition. The salient visual information across images is very useful and plays a significant role for the visually impaired in identifying tactile information. In this paper, we introduce a novel saliency cuts method using local adaptive thresholding to obtain four regions from a given saliency map. First, we produced four regions for image segmentation using a saliency map as an input image and local adaptive thresholding. Second, the four regions were used to initialize an iterative version of the GrabCuts algorithm and to produce a robust and high-quality binary mask with a full resolution. Finally, salient objects’ outer boundaries and inner edges were detected using the solution from our previous research. Experimental results showed that local adaptive thresholding using integral images can produce a more robust binary mask compared to the results from previous works that make use of global thresholding techniques for salient object segmentation. The proposed method can extract salient objects with a low-quality saliency map, achieving a promising performance compared to existing methods. The proposed method has advantages in extracting salient objects and generating simple, important edges from natural scene images efficiently for delivering visually salient information to the visually impaired.

Stereoscopic perception is an important part of human visual system that allows the brain to perceive depth. However, depth information has not been well explored in existing saliency detection models. In this letter, a novel saliency detection method for stereoscopic images is proposed. Firstly, we propose a measure to evaluate the reliability of depth map, and use it to reduce the influence of poor depth map on saliency detection. Then, the input image is represented as a graph, and the depth information is introduced into graph construction. After that, a new definition of compactness using color and depth cues is put forward to compute the compactness saliency map. In order to compensate the detection errors of compactness saliency when the salient regions have similar appearances with background, foreground saliency map is calculated based on depth-refined foreground seeds selection mechanism and multiple cues contrast. Finally, these two saliency maps are integrated into a final saliency map through weighted-sum method according to their importance. Experiments on two publicly available stereo datasets demonstrate that the proposed method performs better than other 10 state-of-the-art approaches.

Transthoracic echocardiography (TTE) is increasingly utilised by point-of-care ultrasound (POCUS) devices. The current study aimed to test whether artificial intelligence (AI) could estimate LVEF using standard TTE without measurements of chamber size or image segmentation while using only parasternal-long-axis and apical-4-chamber views. Methods: Deep learning network training was performed in 76,277 (66%), validated in 17,092 (19%) and tested in 21,920 (15%) randomly selected patients who underwent comprehensive TEE studies between 2007 and 2021. A fourth cohort (N=525) from a different hospital was used for an external validation. A fifth POCUS cohort (N=555) was used for a prospective validation. Accuracy and prognostic significance were evaluated using the hold-out testing cohort. Then, using the same hold-out cohort patients were divided into two groups: (1) patients whose machine-estimated LVEF was ≤ 5% points of the reported value (LVEF discrepancy group [LDG]), (2) all other patients (non-LDG). All-cause mortality and heart failure (HF) hospitalisations were used as the study endpoints. Results: The final hold-out cohort included 21,920 patients with a median age of 69 (IQR 53-79) of whom 12,244 (56%) were men. LVEF was estimated as a continuous variable with a mean average error (MAE) of 3.2, root mean square error (RMSE) of 4.6 and Pearson correlation coefficient (r) of 0.89. Similar results were obtained in both external validation and prospective POCUS cohorts. Among the patients, 2,761 (12.6%) were categorized into the LDG, while 19,159 (87.4%) were classified as non-LDG. During a median follow-up of 6 years, 6,309 (29%) patients died, and 970 (4.4%) patients were admissioned due to HF. A multivariate Cox regression model successfully demonstrated that compared to patients in the non-LDG, individuals in the LDG had an independent and significant 39% and 81% increased risk of death and HF hospitalisation during follow up (95% CI 1.31-1.48, p<0.001 for mortality; 95% CI 1.55-2.12, p<0.001 for HF) (Figure 1). Consistent results were obtained in subgroup analyses examining separately patients with preserved and reduced ejection fraction. An exploratory analysis demonstrated that an abnormality in the regional wall motion of the apical and septal segments had the greatest contribution to the discrepancy between the machine estimated LVEF and reported LVEF (Figure 2A). Consistently, saliency maps have demonstrated that the same areas were captured by the machine as the most contributory areas for LVEF estimation (Figure 2B). Conclusions: Artificial intelligence can estimate LVEF in a method that is agnostic to measurements or image segmentation. We show that echo-based AI-LVEF is an independent predictor of heart-failure hospitalisation and overall survival and, with further evaluation, could be used for risk stratification and estimation of LVEF at the point of care.Figure 1.Graphical abstract Figure 2.Forest plot and saliency maps