deep convolution network
Recently Published Documents

<p>Automated reading of handwritten Kannada documents is highly challenging due to the presence of vowels, consonants and its modifiers. The variable nature of handwriting styles aggravates the complexity of machine based reading of handwritten vowels and consonants. In this paper, our investigation is inclined towards design of a deep convolution network with capsule and routing layers to efficiently recognize  Kannada handwritten characters.  Capsule network architecture is built of an input layer,  two convolution layers, primary capsule, routing capsule layers followed by tri-level dense convolution layer and an output layer.  For experimentation, datasets are collected from more than 100 users for creation of training data samples of about 7769 comprising of 49 classes. Test samples of all the 49 classes are again collected separately from 3 to 5 users creating a total of 245 samples for novel patterns. It is inferred from performance evaluation; a loss of 0.66% is obtained in the classification process and for 43 classes precision of 100% is achieved with an accuracy of 99%. An average accuracy of 95% is achieved for all remaining 6 classes with an average precision of 89%.</p>

As an important imaging modality, mammography is considered to be the global gold standard for early detection of breast cancer. Computer-Aided (CAD) systems have played a crucial role in facilitating quicker diagnostic procedures, which otherwise could take weeks if only radiologists were involved. In some of these CAD systems, breast pectoral segmentation is required for breast region partition from breast pectoral muscle for specific analysis tasks. Therefore, accurate and efficient breast pectoral muscle segmentation frameworks are in high demand. Here, we proposed a novel deep learning framework, which we code-named PeMNet, for breast pectoral muscle segmentation in mammography images. In the proposed PeMNet, we integrated a novel attention module called the Global Channel Attention Module (GCAM), which can effectively improve the segmentation performance of Deeplabv3+ using minimal parameter overheads. In GCAM, channel attention maps (CAMs) are first extracted by concatenating feature maps after paralleled global average pooling and global maximum pooling operation. CAMs are then refined and scaled up by multi-layer perceptron (MLP) for elementwise multiplication with CAMs in next feature level. By iteratively repeating this procedure, the global CAMs (GCAMs) are then formed and multiplied elementwise with final feature maps to lead to final segmentation. By doing so, CAMs in early stages of a deep convolution network can be effectively passed on to later stages of the network and therefore leads to better information usage. The experiments on a merged dataset derived from two datasets, INbreast and OPTIMAM, showed that PeMNet greatly outperformed state-of-the-art methods by achieving an IoU of 97.46%, global pixel accuracy of 99.48%, Dice similarity coefficient of 96.30%, and Jaccard of 93.33%, respectively.

In the current salient object detection network, the most popular method is using U-shape structure. However, the massive number of parameters leads to more consumption of computing and storage resources which are not feasible to deploy on the limited memory device. Some others shallow layer network will not maintain the same accuracy compared with U-shape structure and the deep network structure with more parameters will not converge to a global minimum loss with great speed. To overcome all of these disadvantages, we propose a new deep convolution network architecture with three contributions: (1) using smaller convolution neural networks (CNNs) to compress the model in our improved salient object features compression and reinforcement extraction module (ISFCREM) to reduce parameters of the model. (2) introducing channel attention mechanism to weigh different channels for improving the ability of feature representation. (3) applying a new optimizer to accumulate the long-term gradient information during training to adaptively tune the learning rate. The results demonstrate that the proposed method can compress the model to 1/3 of the original size nearly without losing the accuracy and converging faster and more smoothly on six widely used datasets of salient object detection compared with the others models. Our code is published in https://gitee.com/binzhangbinzhangbin/code-a-novel-attention-based-network-for-fast-salientobject-detection.git

Gait walking patterns are one of the key research topics in natural biometrics. The temporal information of the unique gait sequence of a person is preserved and used as a powerful data for access. Often there is a dive into the flexibility of gait sequence due to unstructured and unnecessary sequences that tail off the necessary sequence constraints. The authors in this work present a novel perspective, which extracts useful gait parameters regarded as independent frames and patterns. These patterns and parameters mark as unique signature for each subject in access authentication. This information extracted learns to identify the patterns associated to form a unique gait signature for each person based on their style, foot pressure, angle of walking, angle of bending, acceleration of walk, and step-by-step distance. These parameters form a unique pattern to plot under unique identity for access authorization. This sanitized data of patterns is further passed to a residual deep convolution network that automatically extracts the hierarchical features of gait pattern signatures. The end layer comprises of a Softmax classifier to classify the final prediction of the subject identity. This state-of-the-art work creates a gait-based access authentication that can be used in highly secured premises. This work was specially designed for Defence Department premises authentication. The authors have achieved an accuracy of 90 % ± 1.3 % in real time. This paper mainly focuses on the assessment of the crucial features of gait patterns and analysis of gait patterns research.

Semantic segmentation, as a pixel-level recognition task, has been widely used in a variety of practical scenes. Most of the existing methods try to improve the performance of the network by fusing the information of high and low layers. This kind of simple concatenation or element-wise addition will lead to the problem of unbalanced fusion and low utilization of inter-level features. To solve this problem, we propose the Inter-Level Feature Balanced Fusion Network (IFBFNet) to guide the inter-level feature fusion towards a more balanced and effective direction. Our overall network architecture is based on the encoder–decoder architecture. In the encoder, we use a relatively deep convolution network to extract rich semantic information. In the decoder, skip-connections are added to connect and fuse low-level spatial features to restore a clearer boundary expression gradually. We add an inter-level feature balanced fusion module to each skip connection. Additionally, to better capture the boundary information, we added a shallower spatial information stream to supplement more spatial information details. Experiments have proved the effectiveness of our module. Our IFBFNet achieved a competitive performance on the Cityscapes dataset with only finely annotated data used for training and has been greatly improved on the baseline network.

Oil tank inventory is significant for the economy and the military, as it can be used to estimate oil reserves. Traditional oil tank detection methods mainly focus on the geometrical characteristics and spectral features of remotely sensed images based on feature engineering. The methods have a limited application capability when the distribution pattern of ground objects in the image changes and the imaging condition varies largely. Therefore, we propose an end-to-end deep convolution network Res2-Unet+, to detect oil tanks in a large-scale area. The Res2-Unet+ method replaces the typical convolution block in the encoder of the original Unet method using hierarchical residual learning branches. A hierarchical branch is used to decompose the feature map into a few sub-channel features. To evaluate the generalization and transferability of the proposed model, we use high spatial resolution images from three different sensors in different areas to train the oil tank detection model. Images from yet another sensor in another area are used to evaluate the trained model. Three more widely used methods, Unet, Segnet, and PSPNet, are trained and evaluated for the same dataset. The experiments prove the effectiveness, strong generalization, and transferability of the proposed Res2-Unet+ method.

As a rare malignant tumor, cervical neuroendocrine cancer (NEC) is difficult in diagnosis even for experienced pathologists. A computer-assisted diagnosis may be helpful for the improvement of diagnostic accuracy. Nevertheless, the computer-aided pathological diagnosis has to face a great challenge that the hundred-million-pixels or even gig-pixels whole slide images (WSIs) cannot be applied directly in the existing deep convolution network for training and analysis. Therefore, the construction of a neural network to realize the automatic screening of cervical NEC is challenging; meanwhile, as far as we know, little attention has been paid to this field. In order to address this problem, here we present a multiple-instance learning method for automatic recognition of cervical NEC on pathological WSI, which consists of the Sliding Detector module and Lesion Analyzer module. A pathological WSI dataset, which is composed of 84 NEC cases and 216 NEC-free cases from the Pathological Department of West China Second University Hospital, is applied to evaluate the performance of the method. The experimental results show that the recall rate, accuracy rate, and precision rate of our method for automatic recognition are 92.9%, 92.7%, and 83.0%, respectively, demonstrating the effectiveness and the potential in clinical practice. The application of this method in computer-assisted pathological diagnosis is expected to decrease the misdiagnosis as well as the false diagnosis of rare cervical NEC, and, consequently, improve the therapeutic effect of cervical cancers.

Biometric systems aim to reliably identify and authenticate an individual using physiological or behavioral characteristics. Traditional systems such as the use of access cards, passwords have shown limitations such as forgotten passwords, stolen cards, etc. As an alternative, biometric systems present themselves as efficient systems with a high reliability due to the physiological characteristics of each individual. This paper focuses on a deep learning method for fingerprint recognition. The described architecture uses a pre-processing phase in which grayscale images are represented on the RGB bands and then merged to obtain color images. On the obtained color images will be extracted the characteristics of the fingerprints textures.The fingerprint images after preprocessing are used in a deep convolution network system for decision making. The method is robust with an accuracy of over 99.43% and 99.53% with the respective variants densenet-201 and ResNet-50.