In the segmentation of skin lesions, there are several difficult phenomena, such as blurred edges, hair occlusion, circular field of vision and diagnostic markers. In response to the above problems, we propose the Gated Axial Transformer with Comprehensive Attention (CA-GAT) to segment skin lesions. First, the U-Net encoder-decoder structure is used as the main framework, and the encoder primarily consists of axial transformer layers. The axial attention mechanism is able to efficiently grasp long-distance dependence while greatly reducing the computational complexity. The gating mechanism allows the model to learn accurate positional encodings without pre-training on large-scale datasets. Second, the triple attention mechanism is introduced into the decoder, thus enabling the model to better differentiate the lesion boundary. Finally, the Local-Global training strategy (LoGo) enables the model to better exclude external interference based on contextual information while improving model performance. We conducted experiments on ISIC2018 dataset. Compared with U-Net, CA-Net, MedT and CA-GAT without LoGo, the Dice coefficient of our model increases by 6.3%, 0.46%, 1.2% and 1.49% respectively, and other indicators are also improved. As indicated by the experiment, the model CA-GAT exhibits favorable segmentation performance.
Semantic segmentation of urban-scale point clouds is widely used in aviation, unmanned aerial vehicles, and autonomous driving. However, owing to the many points in the urban-scale point cloud dataset and massive computation in the learning process, traditional networks often have poor segmentation performance and high costs. In this study, we adopted the Point Transformer network as the baseline and integrated random point sampling and attentive pooling into a new transitiondown block, embedded in the encoder structure of the baseline to improve the speed and accuracy of semantic segmentation. On the challenging SensatUrban dataset, the Point Transformer network and the proposed network obtained mIoU values of 71.1% and 76.8%, respectively. The results show that the proposed network effectively improves the shortcomings of the Point Transformer network and achieves better semantic segmentation performance of urban-scale point clouds.
Determining the EE (End of Expiration) and EI (End of Inspiration) time points in the respiratory cycle is one key step during the 4D image construction from free-breathing dynamic thoracic computed tomography (CT) or magnetic resonance imaging (MRI) acquisitions. However, the cost of manually labeling EE and EI time points is extensive. An automatic image-based EE and EI labeling method makes image annotation independent of the image acquisition process, avoiding use of internal or external markers for the patient during image acquisition. The purpose of this paper is to introduce a novel optical-flow-based technique for finding EE and EI time points from dynamic thoracic MRI acquired during natural tidal-breathing. The diaphragm is tracked as a marker to determine the state of breathing. A region of interest (ROI) containing the diaphragm is selected to calculate the pixel optical flow values between two adjacent time slices. The average optical flow values of all pixels including diaphragm motion speed is used as a reference for labeling EE and EI. When the direction of movement of the diaphragm changes, EE or EI is found depending on the direction of the change. Quantitative evaluation was carried out to evaluate the effectiveness of our method in different locations in the lungs as compared to manual labeling. When tested on 28 patient dynamic thoracic MRI data sets, the average error was found to be less than 1 time point. Automatic labeling greatly shortened the labeling time, requiring less than 8 minutes compared to 4 hours for manual labeling per study.
KEYWORDS: Signal detection, Optical flow, Machine learning, Data modeling, Magnetic resonance imaging, Image acquisition, Lung, Chest, Data acquisition, Feature extraction
4D image construction of thoracic dynamic MRI data provides clinicians the capability of examining the dynamic function of the left and right lungs, left and right diaphragms, and left and right chest wall separately. For the method implemented based on free-breathing rapid 2D slice acquisitions, often part of the acquired data cannot be used for the 4D image reconstruction algorithm because some patients hold their breath or breathe in patterns that differ from regular tidal breathing. Manually eliminating abnormal image slices representing such abnormal breathing is very labor intensive considering that typical acquisitions contain ~3000 slices. This paper presents a novel respiratory signal classification algorithm based on optical flow techniques and a SVM classifier. The optical flow technique is used to track the speed of the diaphragm, and the motion features are extracted to train the SVM classification model. Due to the limited number of abnormal samples usually observed, 118 abnormal signals were generated by simulation by appropriately transforming the normal signals, so that the number of normal and abnormal signals reached 160 and 160, respectively. In the process of model training, our goal is to reduce the error rate of false negative abnormal signal detection (FN) as much as possible even at the cost of increasing false positive misclassification rate (FP) for normal signals. From 10 experiments we conducted, the average FN rate and FP rate reached 5% and 26%, respectively. The accuracy over all (real and simulated) samples was 85%. In all real samples, 82% of the abnormal data were correctly detected.
We report a close connection between the fluctuation characteristics of the electrical derivative
(ED) initial peaks and the 1/f noise intensities of different samples we found during the investigation of
the 1/f noise origins of InGaAs quantum well high-power semiconductor laser diodes (LDs). We conduct contrast measurements on over fifty samples, where the current 1/f noise is measured under
different bias currents, expressed by power spectrum density (PSD) and the EDs are computed from the
current-voltage (I-V) measurement results. Then the influence of 1/f noise on the ED initial peaks is
presented by comparing these parameters of different samples. The results show a clear pattern
between the noise intensity and the ED initial peak fluctuation, and distinct differences between functional and aged LD devices, showing that ED initial peak can also be a non-destructive testing
method for high power LD cavity damage and surface defects.
Currently, unlike IALSC-defined thoracic lymph node zones, no explicitly provided definitions for lymph nodes in other body regions are available. Yet, definitions are critical for standardizing the recognition, delineation, quantification, and reporting of lymphadenopathy in other body regions. Continuing from our previous work in the thorax, this paper proposes a standardized definition of the grouping of pelvic lymph nodes into 10 zones. We subsequently employ our earlier Automatic Anatomy Recognition (AAR) framework designed for body-wide organ modeling, recognition, and delineation to actually implement these zonal definitions where the zones are treated as anatomic objects. First, all 10 zones and key anatomic organs used as anchors are manually delineated under expert supervision for constructing fuzzy anatomy models of the assembly of organs together with the zones. Then, optimal hierarchical arrangement of these objects is constructed for the purpose of achieving the best zonal recognition. For actual localization of the objects, two strategies are used — optimal thresholded search for organs and one-shot method for the zones where the known relationship of the zones to key organs is exploited. Based on 50 computed tomography (CT) image data sets for the pelvic body region and an equal division into training and test subsets, automatic zonal localization within 1–3 voxels is achieved.
An efficient lossless compression scheme for hyperspectral images using conventional recursive least-squares (CRLS) predictor with adaptive prediction bands is proposed. The proposed scheme first calculates the preliminary estimates to form the input vector of the CRLS predictor. Then the number of bands used in prediction is adaptively selected by an exhaustive search for the number that minimizes the prediction residual. Finally, after prediction, the prediction residuals are sent to an adaptive arithmetic coder. Experiments on the newer airborne visible/infrared imaging spectrometer (AVIRIS) images in the consultative committee for space data systems (CCSDS) test set show that the proposed scheme yields an average compression performance of 3.29 (bits/pixel), 5.57 (bits/pixel), and 2.44 (bits/pixel) on the 16-bit calibrated images, the 16-bit uncalibrated images, and the 12-bit uncalibrated images, respectively. Experimental results demonstrate that the proposed scheme obtains compression results very close to clustered differential pulse code modulation-with-adaptive-prediction-length, which achieves best lossless compression performance for AVIRIS images in the CCSDS test set, and outperforms other current state-of-the-art schemes with relatively low computation complexity.
In this paper, we proposed a novel method to extract shape feature based on dual-tree complex wavelet. First, with the two level dual-tree complex wavelet transformations, we can get two low frequency components of the first level, which are used as wavelet moment invariants formed from approximation coefficients. Then, we calculate means and variance for each of the six detailed components in the second level since it contains different directions information of the shape. Using the Principal Component Analysis (PCA), twenty features can be reduced to five maximum useful features which contribute to shape matching.
This paper presents a parallel algorithm designed for 1/f noise signal estimation based on Compressed sensing theory on the GPU platform. In the accelerating process, we select parts of the serial program as the object to be speeded up for the execution time of algorithm. Compared with the conventional methods for 1/f noise estimation, our scheme has shown a 20x speedup.
In this paper, we aimed to separate the 1/f noise from the original signal, and analyzed its characteristics of power spectrum. First, an N-level wavelet transform has been applied to the original data signal before the compressed sensing observation for the original signal. Compared with the tradition measurement procession of compressed sensing, the measurement matrix here is replaced with the circulant matrix. This can greatly reduce the measurement number compared with the random Gaussian matrix. To reduce the algorithm time, some zero independent elements are introduced to the circulant matrix. This proposed circulant matrix is then helpful to save 60 percent of algorithm’s reconstruction time.
In this paper, firstly we established the theoretical model of the relationship between correlation and wavelet vanishing moment of 1/f noise. Then we designed and built the 1/f noise measurement system for semiconductor laser diodes (LDs). Based on this, we introduced the measurement method of low-frequency electrical noise in semiconductor LDs and the extraction process of parameters associated. The wavelet transform decorrelation calculation is applied to the time-domain of the low-frequency noise signals, and the calculation and analysis of the relation between wavelet coefficients variances and scales are completed. The experimental results show that the noise wavelet coefficient correlation E and the vanishing moment N of wavelet function satisfy the decorrelation theoretical model at some scales, which implies that the low-frequency noise signals measured from the LDs belong to 1/f noise. Besides, the wavelet coefficient variance calculation results indicate that the noise measured in the experiments is 1/f noise from another aspect. Finally we theoretically proposed a new idea of semiconductor laser 1/f noise judging based on the methods above.
In order to effectively store and transmit MODIS multispectral data, a lossless compression method based on mix coding
and integer wavelet transform (IWT) is proposed in this paper. Firstly, the algorithm computes the correlation
coefficients between spectrums in MODIS data. Using proper coefficient threshold, the original bands will be divided
two groups: one group use spectral prediction method and then compress residual error, while the other group data is
directly compressed by some standard compressor. For the spectral prediction group, we can find the current band that
has greatest correlation with the previous band by the judgments of correlation coefficient, thus the optimal spectral
prediction sequence is obtained by band reordering. The prediction band data can be computed with the previous band
data and optimal linear predictor, so the spectral redundancy can be eliminated by using spectral prediction. In order to
reduce residual differences in further, the block optimal linear predictor is designed in this paper. Next, except for the
first band of the spectral prediction sequence, the residual errors of other bands are encoded by IWT and SPIHT. The
direct compression bands and the first band of spectral prediction sequence are compressed by JPEG2000. Finally, the
coefficients of block optimal linear predictor and other side information are encoded by adaptive arithmetic coding. The
experimental results show that the proposed method is efficient and practical for MODIS data.
A novel theory of information acquisition-"compressive sampling" has been applied in this paper, and goes against the
common wisdom in data acquisition of Shannon theorem. CS theory asserts that one can recover certain signals and
images perfectly from far fewer samples or measurements than traditional methods use. This paper presents an
improvement on genetic algorithm instead of match pursuit algorithm in consideration of the enormous computational
complexity on sparse decomposition. Then the whole image is divided into small blocks which can be processed by
sparse decomposition, and an end to decomposition is determined by PSNR threshold adaptively. At last, the experiment
results show that good performance on image reconstruction with less computational complexity has been achieved.
Cloud is one of common noises in MODIS remote sensing image. Because of cloud interference, much important
information covered with cloud can't be obtained. In this paper, an effective method is proposed to detect and remove
thin clouds with single MODIS image. The proposed method involves two processing-thin cloud detection and thin
cloud removal. As for thin cloud detection, through analyzing the cloud spectral characters in MODIS thirty-six bands,
we can draw the conclusion that the spectral reflections of ground and cloud are different in various MODIS band.
Hence, the cloud and ground area can be separately identified based on MODIS multispectral analysis. Then, the region
label algorithm is used to label thin clouds from many candidate objects. After cloud detection processing, thin cloud
removal method is used to process each cloud region. Comparing with traditional methods, the proposed method can
realize thin cloud detection and removal with single remote sensing image. Additionally, the cloud removal processing
mainly aims to the cloud label region rather than the whole image, so it can improve the processing efficiency.
Experiment results show the method can effectively remove thin cloud from MODIS image.
A high efficiency and high brightness white LED driver with adaptive current control strategy in 0.5um CMOS process is
presented in this paper. The current control mode is determined by output current level and input voltage value, switching
between normal low drop-out (LDO) regulator and 2×charge-pump mode. The presented whiter LED driver is able to
drive load current as large as 1.1A. The peak efficiency as high as 95% occurs at 220mA load current, with an output
voltage ripple lower than 95mV at 1.1A output current can be achieved.
In this paper, a testing and analyzing system for volt-current and optic-power characteristics of semiconductor laser
diodes (LDs) based on virtual instrumentation is designed and developed by using PCI-6014 DAQ. The design methods
of hardware circuit and software for NI-DAQ are introduced. Some protective methods to LDs, such as driving current
limit, avoiding electrical pulse and delay start-up etc. problems, are completely finished by the software, instead of
mostly considering the resolvent on hardware circuits. The detailed tested data for the function and performance of the
system is presented, and the every tested data of LDs indicates that the whole system is of excellent performance and
stability to obtain the parameters of LDs.
This paper presents a new image mosaic algorithm based on ratio template matching by introducing gradient factor to
enhance the robustness of the template. It corrects the exposure difference in the stitching image, and applies the fade-in
and fade-out method to make the stitched image seamless and smooth. Experimental results testify that this algorithm
improves the reliability and increases the speed of matching and is easier to operate.
A new lossless compression method based on prediction tree with error compensation for hyperspectral imagery is
proposed in this paper. This method incorporates the techniques of prediction tree and adaptive band prediction. The
proposed method is different from previous similar approaches in that its prediction to the current band is performed by
multiple bands and the error created by the prediction tree is compensated by a linear adaptive predictor for decorrelating
spectral statistical redundancy. After de-correlating intraband and interband redundancy, the SPIHT (Set
Partitioning in Hierarchical Trees) wavelet coding is used to encode the residual image. The proposed method achieves
high compression ratio on the NASA JPL AVIRIS data.
The paper discusses the key problem of measurement accuracy in a precise photo electronic measurement system,
through combining the camera calibration method based on computer active vision with digital image processing
technology, a method for calibrating camera internal parameters is proposed in the paper. The optics high accuracy
theodolite and CCD subdivision measurement are combined in the method, and the least squares method is also used to
determined the camera internal parameters under the optimum condition. The experimental results indicate: Compare
with traditional camera calibration method, the operation of the camera calibration method is simple, the calibration
speed is quick, and the applicable scope is broad, it can solve the problem of the distortion of CCD camera well.
Image compression based on lossless or nearly lossless region of interest (ROI) means to lossless compress the
interesting regions and loss compress the uninteresting regions in an image. The technology both may obtain the high
quality image information and maintain the high compression ratio, which solves the contradiction between the image
quality and the image compression ratio. Applying ROI, we can compress image with different accuracy in different
region, which make the important parts of an image be coded with better quality than rest image. In the test project of the
shooting range, a great many target images will be saved, however the tester only is interested in the target region and
not interested the background information. According to the situation, a ROI compress method based on target image is
proposed in the paper. The experiment results show that the method can greatly reduce the image data storage,
meanwhile remain the target information perfectly.
The method of screening semiconductor lasers by using electrical derivative technique is described in detail. The nonlinear diffusion equation is applied to denoising of electrical derivative data according to its denoising theory in signal processing. The denoising experiments of electrical derivative data for several dozens semiconductor lasers indicate that the denoising method can effectively reduce the noise in electrical derivative data and the errors of the measured parameters. The farther experiments indicate that the accurate estimate ratio of the devices can be effectively increased by using the measured parameters which have been denoised, to estimate the quality and reliability of semiconductor lasers.
The main conceptual aspect of this work is on 1/f signal estimate. In this paper a new estimation algorithm that bases on Dual-Tree Complex Wavelet is proposed, which uses the variance of the wavelet coefficients at different scales to estimate the parameters of 1/f process. Adopting Maximum Posteriori estimator estimates the wavelet coefficients of 1/f process. The simulation results show that the method is effective. And comparing with other methods this method doesn't need to know the statistical characteristic of the added white noise and the parameters of the 1/f process.
The shooting range test is an important field in modern weapon development. The modern weaponry is developing towards long distance and automation directions, therefore the shooting range test is put forward new higher requirements. A novel method of target detection based on the digital image processing technology is proposed in the paper. Experiments indicate the strategy is fit to the request of the dynamic target detection and tracking in the shooting range.
In this paper, the index guide and band gap guide polymer microstructured optical fibers are designed. For the index guide fiber, a liquid crystal core is used and 60dB extinction ratio tunable attenuator is obtained. For the bandgap polymer microstructured optical fibers, a regular structure is presented from experiment and an ideal defect can be realized easily by a new method.
Physical vapor deposition in horizontal systems has been used to grow crystal thin-film of organic semiconductor pentacene. Using 10~30mg of starting material, 20~30mm sized crystals thin-film, suitable for characterization measurements or device fabricated, have been grown. And we have measured the sample of pentacene crystal thin-film using TEM. X-ray diffraction electron microscopy. The results indicated that the crystal lattice array order.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.