Cascade recurring deep networks for audible range prediction.

BACKGROUND: Hearing Aids amplify sounds at certain frequencies to help patients, who have hearing loss, to improve the quality of life. Variables affecting hearing improvement include the characteristics of the patients' hearing loss, the characteristics of the hearing aids, and the characteristics of the frequencies. Although the two former characteristics have been studied, there are only limited studies predicting hearing gain, after wearing Hearing Aids, with utilizing all three characteristics. Therefore, we propose a new machine learning algorithm that can present the degree of hearing improvement expected from the wearing of hearing aids. METHODS: The proposed algorithm consists of cascade structure, recurrent structure and deep network structure. For cascade structure, it reflects correlations between frequency bands. For recurrent structure, output variables in one particular network of frequency bands are reused as input variables for other networks. Furthermore, it is of deep network structure with many hidden layers. We denote such networks as cascade recurring deep network where training consists of two phases; cascade phase and tuning phase. RESULTS: When applied to medical records of 2,182 patients treated for hearing loss, the proposed algorithm reduced the error rate by 58% from the other neural networks. CONCLUSIONS: The proposed algorithm is a novel algorithm that can be utilized for signal or sequential data. Clinically, the proposed algorithm can serve as a medical assistance tool that fulfill the patients' satisfaction.

A Map/INS/Wi-Fi Integrated System for Indoor Location-Based Service Applications.

In this research, a new Map/INS/Wi-Fi integrated system for indoor location-based service (LBS) applications based on a cascaded Particle/Kalman filter framework structure is proposed. Two-dimension indoor map information, together with measurements from an inertial measurement unit (IMU) and Received Signal Strength Indicator (RSSI) value, are integrated for estimating positioning information. The main challenge of this research is how to make effective use of various measurements that complement each other in order to obtain an accurate, continuous, and low-cost position solution without increasing the computational burden of the system. Therefore, to eliminate the cumulative drift caused by low-cost IMU sensor errors, the ubiquitous Wi-Fi signal and non-holonomic constraints are rationally used to correct the IMU-derived navigation solution through the extended Kalman Filter (EKF). Moreover, the map-aiding method and map-matching method are innovatively combined to constrain the primary Wi-Fi/IMU-derived position through an Auxiliary Value Particle Filter (AVPF). Different sources of information are incorporated through a cascaded structure EKF/AVPF filter algorithm. Indoor tests show that the proposed method can effectively reduce the accumulation of positioning errors of a stand-alone Inertial Navigation System (INS), and provide a stable, continuous and reliable indoor location service.

Building keypoint mappings on multispectral images by a cascade of classifiers with a resurrection mechanism.

Inspired by the boosting technique for detecting objects, this paper proposes a cascade structure with a resurrection mechanism to establish keypoint mappings on multispectral images. The cascade structure is composed of four steps by utilizing best bin first (BBF), color and intensity distribution of segment (CIDS), global information and the RANSAC process to remove outlier keypoint matchings. Initial keypoint mappings are built with the descriptors associated with keypoints; then, at each step, only a small number of keypoint mappings of a high confidence are classified to be incorrect. The unclassified keypoint mappings will be passed on to subsequent steps for determining whether they are correct. Due to the drawback of a classification rule, some correct keypoint mappings may be misclassified as incorrect at a step. Observing this, we design a resurrection mechanism, so that they will be reconsidered and evaluated by the rules utilized in subsequent steps. Experimental results show that the proposed cascade structure combined with the resurrection mechanism can effectively build more reliable keypoint mappings on multispectral images than existing methods.

Intelligent fault monitoring and diagnosis of tunnel fans using a hierarchical cascade forest.

Tunnel fan is critical fire-fighting equipment, and its safe and stable operation is very important for the efficiency and safety of tunnel traffic. Existing studies commonly train the fault diagnosis methods with the goal of minimizing mean error which ignores the difference between classes in feature distribution. To solve the problem of inaccurate prediction caused by mean error evaluation, this paper presents a non-neural deep learning model, namely hierarchical cascade forest, which has three characteristics: (1) A hierarchical cascade structure is constructed, of which the output comes from each layer; (2) Each fault class is evaluated and recognized independently, the result of fault classes that are easy to distinguish is output earlier; (3) A confidence-based threshold estimate method is proposed in HCF and used to improve the training method to increase the reliability of HCF. Based on these, HCF improves the cascade forest structure and implements the proper matching of different depth of feature and fault patterns. The effect of HCF is verified through experiments based on the tunnel fans testing rig. Experimented results show that, compared to Deep Forest, the accuracy of HCF increases by 0.6% to 10.8%, and the training time of HCF is reduced 33.24%.

Automatic liver tumor segmentation used the cascade multi-scale attention architecture method based on 3D U-Net.

PURPOSE: Due to the complex structure of liver tumors and the low contrast with normal tissues make it still a challenging task to accurately segment liver tumors from CT images. To address these problems, we propose an end-to-end segmentation method for liver tumors. METHODS: The method uses a cascade structure to improve the network's extraction of information. First, the Side-output Feature Fusion Attention block is used to fuse features at different levels and combine with attention mechanism to focus on important information. Then, the Atrous Spatial Pyramid Pooling Attention block is used to extract multi-scale semantic features. Finally, the Multi-scale Prediction Fusion block is used to fully fused the features captured at each layer of the network. RESULT: To verify the performance of the proposed model and the effectiveness of each module, we evaluate it on LiTS and 3DIRCADb datasets and obtained Dice per Case of 0.665 and 0.719, respectively, and Dice Global of 0.812 and 0.784, respectively. CONCLUSION: The proposed method is compared with the basic model 3D U-Net, as well as some mainstream methods based on U-Net variants, and our method achieves better performance on the liver tumor segmentation task and is superior to most segmentation algorithms.

A progressive deep wavelet cascade classification model for epilepsy detection.

Automatic epileptic seizure detection according to EEG recordings is helpful for neurologists to identify an epilepsy occurrence in the initial anti-epileptic treatment. To quickly and accurately detect epilepsy, we proposed a progressive deep wavelet cascade classification model (PDWC) based on the discrete wavelet transform (DWT) and Random Forest (RF). Different from current deep networks, the PDWC mimics the progressive object identification process of human beings with recognition cycles. In every cycle, enhanced wavelet energy features at a specific scale were extracted by DWT and input into a set of cascade RF classifiers to realize one recognition. The recognition accuracy of PDWC is gradually improved by the fusion of classification results produced by multiple recognition cycles. Moreover, the cascade structure of PDWC can be automatically determined by the classification accuracy increment between layers. To verify the performance of the PDWC, we respectively applied five traditional schemes and four deep learning schemes to four public datasets. The results show that the PDWC is not only superior than five traditional schemes, including KNN, Bayes, DT, SVM, and RF, but also better than deep learning methods, i.e. convolutional neural network (CNN), Long Short-Term Memory (LSTM), multi-Grained Cascade Forest (gcForest) and wavelet cascade model (WCM). The mean accuracy of PDWC for all subjects of all datasets reaches to 0.9914. With a flexible structure and less parameters, the PDWC is more suitable for the epilepsy detection of diverse EEG signals.

Domino Effect of Thickness Fluctuation on Subband Structure and Electron Transport within Semiconductor Cascade Structures.

Effectively restraining random fluctuation of layer thickness (RFT) during the thin-film epitaxy plays an essential part in improving the quality of low-dimensional materials for device application. While it is already challenging to obtain an ideal growth condition for thickness control, the tangle of RFT with interfacial problems makes it even more difficult to guarantee the properties of heterostructures and the performance of devices. In our research, the RFT of potential barriers and wells within a semiconductor multilayer is demonstrated to correlate with the interfacial grading effect (IFG) and to affect the band offset strongly. Then, the synergetic effect of RFT and IFG that serves as the first domino is shown to impact the subband structure and the electron transport successively. On the basis of an investigation of a quantum cascade structure, statistical results indicate a normal distribution of RFT with a standard deviation of about 1 Å and an extreme value of 3 Å (about one monolayer) for all the layers within 38 cascade periods. The "seemingly negligible" RFT could actually reduce the conduction band offset for tens to hundreds of meV and alter the subband gaps at a rate of 40 meV/monolayer at most. Furthermore, the dependence of different subband gaps on the barrier/well thickness differs from one another. In addition, the distribution of wave function could also be regulated dramatically by RFT to change the type of electron transition and thus the carrier lifetime. Further impacts of RFT and the RFT-modulated subband alignment on electron transport result in two different mechanisms (injection-dominant and extraction-dominant) of electron population inversion (PI), which is manifested by comparatively discussing the results of in situ electron holography and macro performances.

An effective teeth recognition method using label tree with cascade network structure.

In this article, we apply the deep learning technique to medical field for the teeth detection and classification of dental periapical radiographs, which is important for the medical curing and postmortem identification. We detect teeth in an input X-ray image and distinguish them from different position. An adult usually has 32 teeth, and some of them are similar while others have very different shape. So there are 32 teeth position for us to recognize, which is a challenging task. Convolutional neural network is a popular method to do multi-class detection and classification, but it needs a lot of training data to get a good result if used directly. The lack of data is a common case in medical field due to patients' privacy. In this work, limited to the available data, we propose a new method using label tree to give each tooth several labels and decompose the task, which can deal with the lack of data. Then use cascade network structure to do automatic identification on 32 teeth position, which uses several convolutional neural network as its basic module. Meanwhile, several key strategies are utilized to improve the detection and classification performance. Our method can deal with many complex cases such as X-ray images with tooth loss, decayed tooth and filled tooth, which frequently appear on patients. The experiments on our dataset show: for small training dataset, compared to the precision and recall by training a 33-classes (32 teeth and background) state-of-the-art convolutional neural network directly, the proposed approach reaches a high precision and recall of 95.8% and 96.1% in total, which is a big improvement in such a complex task.

Near-infrared and mid-infrared semiconductor broadband light emitters.

Semiconductor broadband light emitters have emerged as ideal and vital light sources for a range of biomedical sensing/imaging applications, especially for optical coherence tomography systems. Although near-infrared broadband light emitters have found increasingly wide utilization in these imaging applications, the requirement to simultaneously achieve both a high spectral bandwidth and output power is still challenging for such devices. Owing to the relatively weak amplified spontaneous emission, as a consequence of the very short non-radiative carrier lifetime of the inter-subband transitions in quantum cascade structures, it is even more challenging to obtain desirable mid-infrared broadband light emitters. There have been great efforts in the past 20 years to pursue high-efficiency broadband optical gain and very low reflectivity in waveguide structures, which are two key factors determining the performance of broadband light emitters. Here we describe the realization of a high continuous wave light power of >20 mW and broadband width of >130 nm with near-infrared broadband light emitters and the first mid-infrared broadband light emitters operating under continuous wave mode at room temperature by employing a modulation p-doped InGaAs/GaAs quantum dot active region with a 'J'-shape ridge waveguide structure and a quantum cascade active region with a dual-end analogous monolithic integrated tapered waveguide structure, respectively. This work is of great importance to improve the performance of existing near-infrared optical coherence tomography systems and describes a major advance toward reliable and cost-effective mid-infrared imaging and sensing systems, which do not presently exist due to the lack of appropriate low-coherence mid-infrared semiconductor broadband light sources.

Map-Based Indoor Pedestrian Navigation Using an Auxiliary Particle Filter.

In this research, a non-infrastructure-based and low-cost indoor navigation method is proposed through the integration of smartphone built-in microelectromechanical systems (MEMS) sensors and indoor map information using an auxiliary particle filter (APF). A cascade structure Kalman particle filter algorithm is designed to reduce the computational burden and improve the estimation speed of the APF by decreasing its update frequency and the number of particles used in this research. In the lower filter (Kalman filter), zero velocity update and non-holonomic constraints are used to correct the error of the inertial navigation-derived solutions. The innovation of the design lies in the combination of upper filter (particle filter) map-matching and map-aiding methods to further constrain the navigation solutions. This proposed navigation method simplifies indoor positioning and makes it accessible to individual and group users, while guaranteeing the system's accuracy. The availability and accuracy of the proposed algorithm are tested and validated through experiments in various practical scenarios.