CNN-based encoder-decoder networks for salient object detection: A comprehensive review and recent advances

Ji, Yuzhu; Zhang, Haijun; Zhang, Zhao; Liu, Ming

doi:10.1016/j.ins.2020.09.003

Cited by 203 publications

(70 citation statements)

References 92 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where x corresponds to the normalized data obtained from (9). These preprocessing transformations are important for two reasons: 1) the level of energy consumed by DSO's customers varies considerably from one to another, then it is essential to transform the time series to the same scale so they can be comparable; 2) CNN-based methods used in the feature extraction step expect that all features are centered around zero, have variance equal one, and are scaled to range [0,1] previously.…”

Section: ) Datasetsmentioning

confidence: 99%

“…Since there is no trivial way to characterize relevant patterns contained in the raw data, a potential solution is to explore ways to easily accommodate different (and eventually new) kinds of frauds by taking advantage of data-driven strategies. Current Deep Learning (DL) architectures [6] have been demonstrated to be a promising alternative to realize this task, given their remarkable success in several applications, such as medical image analysis [7], financial forecasting [8], salient object detection [9], and even for NTL detection [10]- [14]. A DL method is basically a multilayered artificial neural network that learns hierarchically ways to represent the input data during its training step.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Non-Technical Loss Detection in Power Grid Using Information Retrieval Approaches: A Comparative Study

Esmael

Silva²,

Ji³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Non-technical loss (NTL) detection is a persistent challenge for Distribution System Operators. Data-driven solutions have been widely used nowadays to analyze customers' energy consumption and to identify suspicious fraud patterns for a posterior on-field inspection. However, the usage of such techniques, in particular the current deep learning methods, is not trivial and requires special attention to tackle imbalanced-class and overfitting issues. In this paper, we propose a new non-technical loss detection framework, which combines the effectiveness of convolutional neural network feature extractors with the efficiency of the Information Retrieval paradigm. In our solution, state-of-the-art pre-trained convolution neural networks (CNNs) extract deep features from electricity consumption time series represented as images. Next, these deep features are encoded into textual signatures and indexed using off-the-shelf solutions for posterior fraud searching. With this framework, the user can search for a specific fraud pattern in the utility database without having to train any classifier. The experiments performed in a real dataset provided by CPFL Energia, one of the largest electric utilities in Brazil, presented promising results both in terms of effectiveness and efficiency for the detection of fraudulent customers. In the conducted comparative study, we evaluate different time series image representations and CNN feature extraction approaches with regard to NTL detection results. Experimental results demonstrate that the combination of the Recurrence Plot image representation with the VGG16 CNN presented the best performance in terms of both effectiveness and efficiency.INDEX TERMS Content-based retrieval, deep learning, feature extraction, information retrieval, machine learning, non-technical loss detection, pattern analysis, power grids, and time series retrieval.

show abstract

Section: ) Datasetsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Non-Technical Loss Detection in Power Grid Using Information Retrieval Approaches: A Comparative Study

Esmael

Silva²,

Ji³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Kang et al [30] adopt a progressive learning strategy to learn saliency information and edge information jointly from coarser resolution to finer resolution. Please refer to [48] for more details about the recent SOD methods and fundamental architectures.…”

Section: Related Work a Salient Object Detectionmentioning

confidence: 99%

DANet: Dynamic Salient Object Detection Networks Leveraging Auxiliary Information

et al. 2021

View full text Add to dashboard Cite

Based on Fully Convolutional Networks, recent salient object detection (SOD) methods achieve impressive results. Some studies improve the classical SOD frameworks by utilizing auxiliary information like fixation points, salient numbers and salient edges. These works adopt auxiliary information by embedding sub-networks into the main network. However, how to incorporate auxiliary information regardless of the specific structure with less coupling is unexplored in SOD. In this paper, we present DANet, a new Dynamic network to leverage arbitrary Auxiliary information for SOD. The proposed framework consists of 1) a Dynamic Weight Generator (DWG) which converts arbitrary auxiliary features into dynamic weights, 2) a Dynamic Bridge Block (DBB) which uses dynamic weight convolution to incorporate auxiliary information from DWG and then refines the fused features, and 3) a two-step training strategy to alleviate the side effect caused by drastic changes among different input images. Extensive experiments demonstrate the effect of different auxiliary information and the proposed framework is a universal method to improve SOD with auxiliary information. Comparison experiments show that DANet achieves state-of-the-art performance without any pre-processing and post-processing.

show abstract

“…The advantages of this paper is to extract the most salient informative objects with their respective semantic regions in an image for understanding the whole scene. This extraction simulates the functionality just like the biological visual consideration systems [43]. Therefore, the main motivation is to provide new insights about human biological attentional processes and gives new ways for: understanding the visual attention, complex scene understanding, detecting salient object in a low clutter context, making new artificial intelligence applications and these applications can be based on image or video saliency detection mechanisms [43,11].…”

Section: Introductionmentioning

confidence: 97%

Visualization of Salient Object With Saliency Maps Using Residual Neural Networks

et al. 2021

View full text Add to dashboard Cite

Visual saliency techniques based on Convolutional Neural Networks (CNNs) exhibit an excessive performance for saliency fixation in a scene, but it is harder to train a network in view of their complexity. The imparting Residual Network Model (ResNet) that is more capable to optimize features for predicting salient area in the form of saliency maps within the images. To get saliency maps, an amalgamated framework is presented that contains two streams of Residual Network Model (ResNet-50). Each stream of Reset-50 that is used to enhance the low level and high level semantics features and build a network of 99 layers at two different image scales for generating the normal saliency attention. This model is trained with transfer learning for initialization that is pre-trained on ImageNet for object detection and with some modification to minimize prediction error. At the end, the two streams integrate the features by fusion at low and high scale dimensions of images. This model is fine-tuned on four commonly used datasets and examines both qualitative and quantitative evaluation metrics with state of the art deep saliency models outcomes.

show abstract

CNN-based encoder-decoder networks for salient object detection: A comprehensive review and recent advances

Cited by 203 publications

References 92 publications

Non-Technical Loss Detection in Power Grid Using Information Retrieval Approaches: A Comparative Study

Non-Technical Loss Detection in Power Grid Using Information Retrieval Approaches: A Comparative Study

DANet: Dynamic Salient Object Detection Networks Leveraging Auxiliary Information

Visualization of Salient Object With Saliency Maps Using Residual Neural Networks

Contact Info

Product

Resources

About