Deep learning performs feature extraction through a series of data transformations. Convolutional neural networks (CNNs) are among the most representative methods in deep learning. CNNs enable complex data transformations through convolution operations to extract high-level features. The kernel-based autoencoder (KAE) uses kernelized principal component analysis to perform both linear and nonlinear data transformations. Consequently, KAE provides better interpretability than CNNs in feature extraction. This paper presents a comprehensive study of KAE, including both single-layer and multi-layer configurations. Two distinct multi-layer autoencoder architectures based on the kernel method are proposed. The proposed method is evaluated using handwritten digit images. Experimental results demonstrate the proposed method’s effectiveness in feature extraction. In terms of the structural similarity index measure, the multi-layer kernel-based autoencoder achieves results comparable to those of the convolutional autoencoder. With fewer training samples, the multi-layer kernel-based autoencoder outperforms the convolutional autoencoder.

Monocular visual odometry is an indispensable technique in visual signal tasks, including robotics and navigation. However, the scale ambiguity and poor generalizability of monocular systems remain significant challenges. To address this issue, we propose a novel framework to enhance visual odometry (VO) using monocular depth estimation algorithms and geometry-based methods. Our approach employs a depth point filtering method that utilizes depth information output from multiple depth estimation models with a minimum depth difference strategy. This approach effectively combines depth information from different models, resulting in more accurate scale factor recovery and trajectory prediction using PnP. Experiment results demonstrate that our method outperforms current learning-based VO methods in terms of generalization capability and accuracy. Additionally, we show that incorporating pre-trained models in the VO system to obtain pseudo-labeled depth data significantly enhances the performance of existing geometry-based VO methods.

Monocular depth estimation algorithms aim to explore the possible links between 2D and 3D data, but challenges remain for existing methods to predict consistent depth from a casual video. Relying on camera poses and the optical flow in the time-consuming test-time training phases makes these methods fail in many scenarios and cannot be used for practical applications. In this work, we present a data-driven post-processing method to overcome these challenges and achieve online processing. Based on a deep recurrent network, our method takes the adjacent original and optimized depth map as inputs to learn temporal consistency from the dataset and achieves higher depth accuracy. Our approach can be applied to multiple single-frame depth estimation models and used for various real-world scenes in real-time. In addition, to tackle the lack of a temporally consistent video depth training dataset of dynamic scenes, we propose an approach to generate the training video sequences dataset from a single image based on inferring motion field. To the best of our knowledge, this is the first data-driven plug-and-play method to improve the temporal consistency of depth estimation for casual videos. Extensive experiments on three datasets and three depth estimation models show that our method outperforms the state-of-the-art methods.

This paper presents a data gathering protocol for wireless sensor network applications that require high throughput, reliability, and low-power. Insofar as high throughput is concerned, TDMA is more suitable than CSMA. Efficient TDMA protocols require solid synchronization and more optimized time slot scheduling; reliable data collection calls for end-to-end packet retransmission. Such factors hinder the implementation of a comprehensive data collection protocol. In this paper, a heuristic TDMA scheduling algorithm is provided to achieve optimized time slot scheduling with multi-channel communication. The combination of TDMA and multi-channel communication affords to sustain high throughput and low-power. In addition, constructive interference based flooding is adopted to provide robust synchronization and efficient downward communication. We associate constructive interference based flooding with packet retransmission to implement collection reliability. The proposed protocol is implemented on Tmote Sky sensor node with Contiki operating system. Experimental results in a deployed network consisting of 29 sensor nodes show that it is able to provide throughput as high as 14.66KByte/s with 100% data collection.

BACKGROUND: Screening of the brain computerised tomography (CT) images is a primary method currently used for initial detection of patients with brain trauma or other conditions. In recent years, deep learning technique has shown remarkable advantages in the clinical practice. Researchers have attempted to use deep learning methods to detect brain diseases from CT images. Methods often used to detect diseases choose images with visible lesions from full-slice brain CT scans, which need to be labelled by doctors. This is an inaccurate method because doctors detect brain disease from a full sequence scan of CT images and one patient may have multiple concurrent conditions in practice. The method cannot take into account the dependencies between the slices and the causal relationships among various brain diseases. Moreover, labelling images slice by slice spends much time and expense. Detecting multiple diseases from full slice brain CT images is, therefore, an important research subject with practical implications. RESULTS: In this paper, we propose a model called the slice dependencies learning model (SDLM). It learns image features from a series of variable length brain CT images and slice dependencies between different slices in a set of images to predict abnormalities. The model is necessary to only label the disease reflected in the full-slice brain scan. We use the CQ500 dataset to evaluate our proposed model, which contains 1194 full sets of CT scans from a total of 491 subjects. Each set of data from one subject contains scans with one to eight different slice thicknesses and various diseases that are captured in a range of 30 to 396 slices in a set. The evaluation results present that the precision is 67.57%, the recall is 61.04%, the F1 score is 0.6412, and the areas under the receiver operating characteristic curves (AUCs) is 0.8934. CONCLUSION: The proposed model is a new architecture that uses a full-slice brain CT scan for multi-label classification, unlike the traditional methods which only classify the brain images at the slice level. It has great potential for application to multi-label detection problems, especially with regard to the brain CT images.

Cataract is one of the most common causes of visual blindness, about 90% of the elderly over 60 years old with visual impairment in China have cataract diseases, and about 90% of eye diseases are diagnosed by observing the fundus. The observation of fundus has always been a necessary mean of diagnosing a cataract. Moreover, it is highly uncertain about judging the degree of lesions based on experience, but also the efficiency of this method is very low. Therefore, employing a computer-aided diagnostic system to perform the automatic grading of cataract is of great research value of practical use. Most of the studies reported in the literature utilize histogram equalization (Histeq) or other image enhancement methods based on gray value changes to improve the contrast. In this paper, the adaptive window model (AWM) is used to enhance the contrast between the vessel and the background. We used features extracted from the spoke features of the image for cataract grading. The best average accuracy achieved by Support Vector Machine (Back Propagation Neural Network) is 80.12% (78.26%) when AWM is used to enhance the contrast. Furthermore, it is even higher than 73.29% (75.16%) when the Histeq is used as an image enhancement technique.

In this paper, a novel Takagi-Sugeno (T-S) fuzzy-system-based model is proposed for hysteresis in piezoelectric actuators. The antecedent and consequent structures of the developed fuzzy hysteresis model (FHM) can be identified online through uniform partition approach and recursive least squares (RLS) algorithm, respectively. With respect to the controller design, the inverse of FHM is used to develop a fuzzy internal model (FIM) controller. Decreasing the hysteresis effect, the FIM controller has a good performance of high-speed trajectory tracking. To achieve nanometer-scale tracking precision, the novel fuzzy adaptive internal model (FAIM) controller is uniquely developed. Based on real-time input and output data to update FHM, the FAIM controller is capable of compensating for the hysteresis effect of the piezoelectric actuator in real time. Finally, the experimental results for two cases are shown: the first is with 50 Hz and the other with multiple-frequency (50 + 25 Hz) sinusoidal trajectories tracking that demonstrate the efficiency of the proposed controllers. Especially, being 0.32% of the maximum desired displacement, the maximum error of 50-Hz sinusoidal tracking is greatly reduced to 6 nm. This result clearly indicates the nanometer-scale tracking performance of the novel FAIM controller.

In this paper, a new type of decoupled 2-DOF translational parallel micro-positioning stage is designed to realize the 2-DOF ultra-precision linear motion. The stage consists of two piezoelectric actuators (PZTs) and a monolithic compliant mechanism. The monolithic compliant mechanism adopts two types of compound double parallel four-leaf flexures and a mirror symmetric structure to reduce the input and output cross coupling and parasitic motion. Based on the stiffness matrix method and screw theory, a mathematical model is constructed to analyze the compliant mechanism. The optimal design is performed in view of performance constraints. The design results show good static and dynamic performances of the stage, which are well validated by the simulation of finite-element-analysis (FEA) and experimental results. The experimental results show that the proposed stage has a full range of 40µm × 40µm when the full voltage(100V) is applied on the two PZTs. Besides, the stage only has the maximum cross coupling of -50dB between the two axes, low enough to utilize single-input-single-out(SISO) control strategies for positioning and tracking.