Detection of urine sediment microscopic images of human urine samples plays an important part in vitro examination. Doctors usually use automatic urine sediment analyzer to assist manual examine. At present, automatic urine sediment analyzers mostly use traditional method of artificial feature extraction to recognize urine sediment images. However, traditional image processing methods based on the selection and combination of feature operators and classifiers require a lot of work and subjective experience for engineers in the implementation process. It's also difficult to deal with urine sediment images recognition tasks with large scale categories, and particles in some different categories are often confused in recognition using traditional image processing methods, such as red blood cells (RBCs) and white blood cells (WBCs). In this paper, a combination convolution neural network (CNN) recognition method with area feature algorithm is proposed. The disadvantage that CNN can weaken the area feature of input image is solved by area feature algorithm (AFA) proposed in this paper. The network models which use 300,000 urine sediment images for training can quickly and accurately recognize 10 categories of urine sediment images, and several confusing categories' recognition indexes are remarkably improved. The test accuracy in the test set reached 97%. INDEX TERMS Automatic diagnosis, deep learning, image processing, urine sediment.
With the increasing application of computer vision technology in autonomous driving, robot, and other mobile devices, more and more attention has been paid to the implementation of target detection and tracking algorithms on embedded platforms. The real-time performance and robustness of algorithms are two hot research topics and challenges in this field. In order to solve the problems of poor real-time tracking performance of embedded systems using convolutional neural networks and low robustness of tracking algorithms for complex scenes, this paper proposes a fast and accurate real-time video detection and tracking algorithm suitable for embedded systems. The algorithm combines the object detection model of single-shot multibox detection in deep convolution networks and the kernel correlation filters tracking algorithm, what is more, it accelerates the single-shot multibox detection model using field-programmable gate arrays, which satisfies the real-time performance of the algorithm on the embedded platform. To solve the problem of model contamination after the kernel correlation filters algorithm fails to track in complex scenes, an improvement in the validity detection mechanism of tracking results is proposed that solves the problem of the traditional kernel correlation filters algorithm not being able to robustly track for a long time. In order to solve the problem that the missed rate of the single-shot multibox detection model is high under the conditions of motion blur or illumination variation, a strategy to reduce missed rate is proposed that effectively reduces the missed detection. The experimental results on the embedded platform show that the algorithm can achieve real-time tracking of the object in the video and can automatically reposition the object to continue tracking after the object tracking fails.
An infrared (IR) sub-imaging system is composed of an optical scanning device and a single IR detector, which provides the target location information to the servo system. Currently, further improvement of positioning accuracy and imaging quality in the traditional rosette scanning guidance mode is experiencing a bottleneck. The emergence of the compressed sensing (CS) technique provides a new solution for this problem as it can recover a high-resolution IR image including richer information with fewer sampling points. In this paper, the complementarity of the CS framework and IR rosette sub-imaging system was analyzed. A new method to improve the resolution of reconstructed IR images, multi-frame joint compressive imaging (MJCI), was proposed. The simulation results revealed the potential of the CS technique when applied to the IR sub-imaging system and demonstrated that the proposed method performed well for reconstruction.
Software defined radios typically employ digital filter banks (FBs) for channelisation. Frequency response masking (FRM) approach is used to reduce the computational complexity of the complex modulated FB with narrow transition band. A unified FRM-based complex modulated FB structure is proposed to design sharp filters with arbitrary passband bandwidth. The proposed unified FRM-based complex modulated FB structure with low complexity is suitable for different even-stacked or odd-stacked, maximally decimated or non-maximally decimated structures. It also can be directly applied to high sampling rate systems. It is shown that the proposed unified FRM-based complex modulated FB structure can obtain the improvement of the computational complexity compared with the other methods.
A spatial distribution of an image that retains all of the multicomponent sample information in the spectral channels can be obtained using multivariate analysis methods, such as principal component analysis (PCA). Most multivariate methods build classifiers based on the spectral features extracted from high-dimensional space. However, such mathematical models have been devoted to exploring the spectral features contained in full spectrum bands, and lack the chemical specificity of the mid-infrared spectrum. In this report, we present the results of a novel characteristic absorption peak interval (CAPI) method to extract spectral band characteristics. This CAPI method extracts absorption peak bands from the spectral dimension by implementing four developed strategies of subspace partition (SP), thereby capturing the subspace characteristic information in multiple adjacent functional group areas. Finally, stacked absorption peak bands are utilized to obtain more efficient distribution information from the multiple subspace feature sets, and then an extreme learning machine is used to perform classification. Experimental results show that the proposed all-sub-band randomization strategy, subspace randomization strategy, probabilistic principal component analysis of the subspace partition, and optimum index factor of the subspace partition are all effective for mid-infrared spectroscopy microscopic image classification. Compared to PCA, 2D principal component analysis, probabilistic principal component analysis, kernel principal component analysis, and PCA linear discriminant analysis, experimental results for three Fourier transform infrared spectroscopy (FTIR) microscopy imaging datasets show that the proposed CAPI method outperforms the other multivariate methods, achieves a higher overall and average classification accuracy, and retains the physical meaning of the spectrum.
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.