Standard Convolutional Layers Research Articles

Compact and Accurate Scene Text Detector

Scene text detection is the task of detecting word boxes in given images. The accuracy of text detection has been greatly elevated using deep learning models, especially convolutional neural networks. Previous studies commonly aimed at developing more accurate models, but their models became computationally heavy and worse in efficiency. In this paper, we propose a new efficient model for text detection. The proposed model, namely Compact and Accurate Scene Text detector (CAST), consists of MobileNetV2 as a backbone and balanced decoder. Unlike previous studies that used standard convolutional layers as a decoder, we carefully design a balanced decoder. Through experiments with three well-known datasets, we then demonstrated that the balanced decoder and the proposed CAST are efficient and effective. The CAST was about 1.1x worse in terms of the F1 score, but 30∼115x better in terms of floating-point operations per second (FLOPS).

DENSE-INception U-net for medical image segmentation

Background and objectiveConvolutional neural networks (CNNs) play an important role in the field of medical image segmentation. Among many kinds of CNNs, the U-net architecture is one of the most famous fully convolutional network architectures for medical semantic segmentation tasks. Recent work shows that the U-net network can be substantially deeper thus resulting in improved performance on segmentation tasks. Though adding more layers directly into network is a popular way to make a network deeper, it may lead to gradient vanishing or redundant computation during training. MethodsA novel CNN architecture is proposed that integrates the Inception-Res module and densely connecting convolutional module into the U-net architecture. The proposed network model consists of the following parts: firstly, the Inception-Res block is designed to increase the width of the network by replacing the standard convolutional layers; secondly, the Dense-Inception block is designed to extract features and make the network more deep without additional parameters; thirdly, the down-sampling block is adopted to reduce the size of feature maps to accelerate learning and the up-sampling block is used to resize the feature maps. ResultsThe proposed model is tested on images of blood vessel segmentations from retina images, the lung segmentation of CT Data from the benchmark Kaggle datasets and the MRI scan brain tumor segmentation datasets from MICCAI BraTS 2017. The experimental results show that the proposed method can provide better performance on these two tasks compared with the state-of-the-art algorithms. The results reach an average Dice score of 0.9857 in the lung segmentation. For the blood vessel segmentation, the results reach an average Dice score of 0.9582. For the brain tumor segmentation, the results reach an average Dice score of 0.9867. ConclusionsThe experiments highlighted that combining the inception module with dense connections in the U-Net architecture is a promising approach for semantic medical image segmentation.

Lightweight efficient network for defect classification of polarizers

SummaryIn industrial production, detecting polarizer defects online and in real time is necessary. Existing methods of detecting polarizer defects based on deep learning can ensure the accuracy of classification; however, there are several issues associated with these methods. These include the models having low detection speed, consuming large amounts of memory, and being difficult to be transplanted into online detection systems. To solve the aforementioned problems, a lightweight efficient network (LWEN) structure based on deep learning was designed, which improves the standard convolution layer and the fully connected (FC) layer to minimize the training model size and increase the speed of classification without reducing the accuracy of classification. First, a new building block, the shunt module, was designed to build the LWEN. Subsequently, a global average pooling layer was used to reduce the spatial resolution to 1 before the FC layer. These key technologies were designed to reduce the number of network parameters and minimize the model size of the network. Experimental results show that the proposed LWEN outperforms the state‐of‐the‐art approaches in terms of classification accuracy, speed, and model size.

Implicit and Explicit Concept Relations in Deep Neural Networks for Multi-Label Video/Image Annotation

In this paper, we propose a deep convolutional neural network (DCNN) architecture that addresses the problem of video/image concept annotation by exploiting concept relations at two different levels. At the first level, we build on ideas from multi-task learning, and propose an approach to learn concept-specific representations that are sparse, linear combinations of representations of latent concepts. By enforcing the sharing of the latent concept representations, we exploit the implicit relations between the target concepts. At a second level, we build on ideas from structured output learning and propose the introduction, at training time, of a new cost term that explicitly models the correlations between the concepts. By doing so, we explicitly model the structure in the output space (i.e., the concept labels). Both of the above are implemented using standard convolutional layers and are incorporated in a single DCNN architecture that can then be trained end-to-end with standard back-propagation. Experiments on four large-scale video and image data sets show that the proposed DCNN improves concept annotation accuracy and outperforms the related state-of-the-art methods.

Adapted deep convnets technology for robust iris recognition

Iris recognition (IR) is widely adopted for human identification in the biometric fields. However, results using handcrafted features designed with the domain knowledge are obtained by conventional IR algorithms which are complex. Convolutional neural networks (convnets) have been proved to be robust with superior learning capability in many computer vision fields. The deep convnets perform better than shallower convnets. However, training the deep convnets for IR is difficult. IrisConvDeeper, an adaptive architecture based on convnets combining Softmax classifier that is used to alleviate the problems of training deep convnets for IR, is proposed. In addition, a better performance is obtained. Furthermore, a shallower architecture termed as IrisConvShallower is designed for comparison. IrisConvDeeper consists of structures called dense block and has 12 or 14 convolutional layers. Meanwhile, IrisConvShallower consists of six or seven standard convolutional layers. The performance of the proposed architectures is tested on two public iris databases, which are collected under different conditions: CASIA-Iris-V3 and IITD iris image databases. The experimental results demonstrate that the proposed IrisConvDeeper outperforms most of state-of-the-art approaches in terms of correct recognition rate, except some approaches that adopted specific methods and identified fewer subjects compared with our methods.

Scene Classification from Synthetic Aperture Radar Images Using Generalized Compact Channel-Boosted High-Order Orderless Pooling Network

The convolutional neural network (CNN) has achieved great success in the field of scene classification. Nevertheless, strong spatial information in CNN and irregular repetitive patterns in synthetic aperture radar (SAR) images make the feature descriptors less discriminative for scene classification. Aiming at providing more discriminative feature representations for SAR scene classification, a generalized compact channel-boosted high-order orderless pooling network (GCCH) is proposed. The GCCH network includes four parts, namely the standard convolution layer, second-order generalized layer, squeeze and excitation block, and the compact high-order generalized orderless pooling layer. Here, all of the layers are trained by back-propagation, and the parameters enable end-to-end optimization. First of all, the second-order orderless feature representation is acquired by the parameterized locality constrained affine subspace coding (LASC) in the second-order generalized layer, which cascades the first and second-order orderless feature descriptors of the output of the standard convolution layer. Subsequently, the squeeze and excitation block is employed to learn the channel information of parameterized LASC statistic representation by explicitly modelling interdependencies between channels. Lastly, the compact high-order orderless feature descriptors can be learned by the kernelled outer product automatically, which enables low-dimensional but highly discriminative feature descriptors. For validation and comparison, we conducted extensive experiments into the SAR scene classification dataset from TerraSAR-X images. Experimental results illustrate that the GCCH network achieves more competitive performance than the state-of-art network in the SAR image scene classification task.

Single Image Super-Resolution Using Dual-Branch Convolutional Neural Network

Recent advances in convolutional neural networks have demonstrated impressive reconstruction for single image super-resolution (SR). However, most of the existing methods are to achieve better performance by directly stacking more convolution layers in a chain way. As the depth increases, it not only lacks time efficiency but also consumes more computer memory. In this paper, we propose a low-complexity dual-branch network, termed DBCN, to deal with the SR problem. The feature extraction part of the proposed DBCN mainly consists of two parallel sub-networks, one extracts fine local information via standard convolution layers, and the other focuses on more contextual information in larger regions by using dilated convolution layers. In addition, deconvolution layers are integrated into the network after feature extraction stage to accelerate the upscaling process. Specifically, a novel skip connection is proposed to learn residual mapping and ease the training. Extensive evaluations on benchmark datasets show the effectiveness of our algorithm, achieving comparable results in terms of accuracy and visual quality compared with several state-of-the-art methods.

A High-Efficiency Fully Convolutional Networks for Pixel-Wise Surface Defect Detection

In this paper, we propose a highly efficient deep learning-based method for pixel-wise surface defect segmentation algorithm in machine vision. Our method is composed of a segmentation stage (stage 1), a detection stage (stage 2), and a matting stage (stage 3). In the segmentation stage, a lightweight fully convolutional network (FCN) is employed to make a pixel-wise prediction of the defect areas. Those predicted defect areas act as the initialization of stage 2, guiding the process of detection to correct the improper segmentation. In the matting stage, a guided filter is utilized to refine the contour of the defect area to reflect the real abnormal region. Besides that, aiming to achieve the tradeoff between efficiency and accuracy, and simultaneously we use depthwise&pointwise convolution layer, strided depthwise convolution layer, and upsample depthwise convolution layer to replace the standard convolution layer, pooling layer, and deconvolution layer, respectively. We validate our findings by analyzing the performance obtained on the dataset of DAGM 2007.

SKFlow: Optical Flow Estimation Using Selective Kernel Networks

Leveraging on the recent developments in convolutional neural networks (CNNs), optical flow estimation from adjacent frames has been cast as a learning problem, with performance exceeding traditional approaches. The existing networks always use standard convolutional layers for extracting multi-level features with the fixed kernel size at each level. For enlarging the receptive field, some works introduce dilated convolution operation, which can capture more contextual information and can avoid the loss of motion details. However, these networks lack the ability to adaptively adjust its receptive field size and cannot aggregate multi-scale information with a selective mechanism. To address this problem, in this paper, we introduce selective kernel network into optical flow estimation, which can adaptively select different scale features and adjust their receptive field according to the global information. Specifically, we conduct the selective kernel mechanism on each level of pyramid, which can adaptively select multi-scale feature at each pyramidal level. The extensive analyses are conducted on MPI-Sintel and KITTI datasets to verify the effectiveness of the proposed approach. The experimental results show that our model achieves comparable results with the previous state-of-the-art networks while keeping a small model size.

Multi-focus image fusion using deep support value convolutional neural network

A novel multi-focus image fusion algorithm based on deep support value convolutional neural network (DSVCNN) is proposed for multi-focus image fusion. First, a deep support value training network is presented by replacing the empirical risk minimization-based loss function by a loss function based on structural risk minimization during the training of convolutional neural network (CNN). Then, to avoid the loss of information, max-pooling/subsampling of the feature mapping layer of a conventional convolutional neural network, which is employed in all conventional CNN frameworks to reduce the dimensionality of the feature map, is replaced by standard convolutional layers with a stride of two. The experimental results demonstrate that the suggested DSVCNN-based method is competitive with current state-of-the-art approaches and superior to those that use traditional CNN methods.

Deformable Faster R-CNN with Aggregating Multi-Layer Features for Partially Occluded Object Detection in Optical Remote Sensing Images

The region-based convolutional networks have shown their remarkable ability for object detection in optical remote sensing images. However, the standard CNNs are inherently limited to model geometric transformations due to the fixed geometric structures in its building modules. To address this, we introduce a new module named deformable convolution that is integrated into the prevailing Faster R-CNN. By adding 2D offsets to the regular sampling grid in the standard convolution, it learns the augmenting spatial sampling locations in the modules from target tasks without additional supervision. In our work, a deformable Faster R-CNN is constructed by substituting the standard convolution layer with a deformable convolution layer in the last network stage. Besides, top-down and skip connections are adopted to produce a single high-level feature map of a fine resolution, on which the predictions are to be made. To make the model robust to occlusion, a simple yet effective data augmentation technique is proposed for training the convolutional neural network. Experimental results show that our deformable Faster R-CNN improves the mean average precision by a large margin on the SORSI and HRRS dataset.

DRINet for Medical Image Segmentation.

Convolutional neural networks (CNNs) have revolutionized medical image analysis over the past few years. The U-Net architecture is one of the most well-known CNN architectures for semantic segmentation and has achieved remarkable successes in many different medical image segmentation applications. The U-Net architecture consists of standard convolution layers, pooling layers, and upsampling layers. These convolution layers learn representative features of input images and construct segmentations based on the features. However, the features learned by standard convolution layers are not distinctive when the differences among different categories are subtle in terms of intensity, location, shape, and size. In this paper, we propose a novel CNN architecture, called Dense-Res-Inception Net (DRINet), which addresses this challenging problem. The proposed DRINet consists of three blocks, namely a convolutional block with dense connections, a deconvolutional block with residual inception modules, and an unpooling block. Our proposed architecture outperforms the U-Net in three different challenging applications, namely multi-class segmentation of cerebrospinal fluid on brain CT images, multi-organ segmentation on abdominal CT images, and multi-class brain tumor segmentation on MR images.