In recent years, deep learning has been widely applied for mammographic image classification. However, most of the existing methods are based on a single mammography view and cannot sufficiently extract discriminative features, thereby resulting in an unsatisfactory classification accuracy. To solve this problem and improve the mammographic image classification performance, we propose a novel multi-view convolutional neural network (CNN) based on multiple mammography views in this paper. Considering that the images acquired from different perspectives contain different and complementary breast mass information, we modify the CNN architecture to exploit the complementary information from the various views of mammography. The new architecture can extract discriminative features from the mediolateral oblique (MLO) and craniocaudal (CC) views of the mammographic images and can effectively incorporate these features for mammographic images. The dilated convolutional layers enable the feature maps extracted from the multiple breast mass views to capture information from a large ''field of vision''. Moreover, multiscale features are obtained by using the convolutional and dilated convolutions. In addition, we incorporate a penalty term into the cross entropy loss function, which enables the model evolution to reduce the misclassification rate by enhancing the contributions of the samples misclassified in the training process. The proposed method was evaluated and compared with several state-of-the-art methods on the open Digital Database for Screening Mammography (DDSM) and Mammographic Image Analysis Society (MIAS) datasets. The experimental results show that the proposed method exhibits a better performance than those of the state-of-the-art methods.INDEX TERMS Medical image processing, mammographic image, deep learning, convolutional neural network.
License plate detection and recognition are still important and challenging tasks in natural scenes. At present, most methods have favorable effect on license plate recognition under restrictive conditions, and most of such license plates are shot under good angle and light conditions. However, for license plates under non-restrictive conditions, such as dark, bright, rotated conditions etc. from the Chinese City Parking Dataset (CCPD), the performance of some methods of license plate recognition will be significantly reduced. In order to improve the accuracy of license plate recognition under unrestricted conditions, a robust license plate recognition model is proposed in this paper, which mainly includes license plate feature extraction, license plate character localization, and feature extraction of characters. First of all, the model can activate the regional features of characters and fully extract the character features of license plates. Then locate each license plate character through Bi-LSTM combined with the context location information of license plates. Finally, 1D-Attention is adopted to enhance useful character features after Bi-LSTM positioning, and reduce useless character features to realize effective acquisition of character features of license plates. A large number of experimental results demonstrate that the proposed algorithm has good performance under unrestricted conditions, which proves the effectiveness and robustness of the model. In CCPD-Base, CCPD-DB, CCPD-FN, CCPD-Tilt, CCPD-Weather, CCPD-Challenge and other sub-datasets, the recognition rates reach 99.3%, 98.5%, 98.6%, 96.4%, 99.3% and 86.6% respectively.
Palmprint recognition has received tremendous research interests due to its outstanding user-friendliness such as non-invasive and good hygiene properties. Most recent palmprint recognition studies such as deep-learning methods usually learn discriminative features from palmprint images, which usually require a large number of labeled samples to achieve a reasonable good recognition performance. However, palmprint images are usually limited because it is relative difficult to collect enough palmprint samples, making most existing deep-learning-based methods ineffective. In this paper, we propose a heuristic palmprint recognition method by extracting triple types of palmprint features without requiring any training samples. We first extract the most important inherent features of a palmprint, including the texture, gradient and direction features, and encode them into triple-type feature codes. Then, we use the block-wise histograms of the triple-type feature codes to form the triple feature descriptors for palmprint representation. Finally, we employ a weighted matching-score level fusion to calculate the similarity between two compared palmprint images of triple-type feature descriptors for palmprint recognition. Extensive experimental results on the three widely used palmprint databases clearly show the promising effectiveness of the proposed method.
The semantic segmentation of remote sensing (RS) image is a hot research field. With the development of deep learning, the semantic segmentation based on a full convolution neural network greatly improves the segmentation accuracy. The amount of information on the RS image is very large, but the sample size is extremely uneven. Therefore, even the common network can segment RS images to a certain extent, but the segmentation accuracy can still be greatly improved. The common neural network deepens the network to improve the classification accuracy, but it has a lot of loss to the target spatial features and scale features, and the existing common feature fusion methods can only solve some problems. A segmentation network is built to solve the above problems very well. The network employs the InceptionV-4 network as the backbone and improves it. We modify the network structure and introduce the changed Atrous Spatial Pyramid Pooling module to extract the multi-scale features of the target from different training stages. Without losing the depth of the network, using Inception blocks to strengthen the width of the network can obtain more abstract features. At the same time, the backbone network is used for semantic fusion of the context, it can retain more spatial features, then an effective decoder network is designed. Finally, evaluate our model on the ISPRS 2D Semantic Labeling Contest Potsdam and Inria Aerial Image Labeling Dataset. The results show that the network has very superior performance, reaching 89.62% IOU score and 94.49% F1 score on the Potsdam dataset, and the IOU score on the Inria dataset has been greatly improved.
The remote sensing (RS) images are widely used in various industries, among which semantic segmentation of RS images is a common research direction. At the same time, because of the complexity of target information and the high similarity of features between the classes, this task is very challenging. In recent years, semantic segmentation algorithms of RS images have emerged in an endless stream, but most of them are improved around the scale features of the target, and the accuracy has great room for improvement. In this case, we propose a semantic segmentation framework for RS images with dynamic perceptual loss. The framework is improved based on the InceptionV-4 network to form a network that includes contextual semantic fusion and dual-channel atrous spatial pyramid pooling (ASPP). The semantic segmentation network is an encoder-decoder structure. In addition, we design a dynamic perceptual loss module and a dynamic loss fusion strategy by further observing the loss changes of the network, so as to better improve the classified details. Finally, experiment on the ISPRS 2D Semantic Labeling Contest Vaihingen Dataset and Massachusetts Building Dataset. Compared with some segmentation networks, our model has excellent performance.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.