Deep learning has become a powerful tool to automatically classify medical hyperspectral images (MedHSIs) for the diagnosis of various tumors such as cancer. These deep based classification approaches consist both feature extraction and disease prediction, which are independent each other. Therefore, extracted features may be incompatible with used classifier prediction. To remedy deficiency, in this work, we propose novel method, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML"...

This article presents a novel deep network with irregular convolutional kernels and self-expressive property (DIKS) for the classification of hyperspectral images (HSIs). Specifically, we use principal component analysis (PCA) superpixel segmentation to obtain series patches, which are regarded as our network. With such kernels, feature maps HSIs can be adaptively computed well describe characteristics each object class. After multiple layers, features exported by all convolution operations...

Hyperspectral image (HSI) classification is a current research hotspot. Most existing methods usually export discriminative features with low-quality distribution and low information utilization, which may induce performance degeneration. To remedy such deficiencies, we propose diagonalized low-rank learning (DLRL) model for HSI in this study. Specifically, classwise regularization used to capture the block-diagonal structure of representation, can further cluster represented pixels from one...

With the rapid development of urbanization, ambient air pollution is becoming increasingly serious. Out many pollutants, fine particulate matter (PM2.5) pollutant that affects urban atmospheric environment to greatest extent. Fine concentration prediction great significance human health and environmental protection. This paper proposes a CNN-SSA-DBiLSTM-attention deep learning fusion model. took meteorological observation data from eight stations in Bijie 1 January 2015 31 December 2022 as...

The paper suggests a new modified approach to improve the performance of particle swarm optimization (PSO). Inspired by intelligent behaviors natural biotic populations, PSO is based on an adaptive strategy, should stop inertia movement enhance learning from its experiences and neighbors when it found be in wrong searching direction, process fly straight nearest destination swarm. Furthermore, four different forms model are presented. Comparison results with standard examination some...

A new solving method based on information fusion estimation (IFE) for linear quadratic optimal control problem is proposed in this paper. According to theory, by fusing the hard constraint of system dynamic equation and soft desired state/output from performance index function, an costate sequence achieved. By law obtained. Based method, state regulator tracker are deduced, computational results which identical traditional methods through theoretical analysis.

In this presented work, a new tracking control strategy is designed for kind of nonholonomic system under external disturbances. First, the whole divided into two design stages by using relay switching technique. Then, state transformations, primitive subsystems are converted unconstrained systems which controllers based on dynamic surface control. Different from most existing finite-time strategies and fixed-time strategies, our able to error within specified accuracy time, starting...

The algal limestone reservoir has extremely low permeability, developed dissolution pores and weak structural planes, heterogeneity. There is a problem of significant differences in single-well production after fracturing oilfield sites. It crucial to clarify the matching relationship between hydraulic fracture parameters production. This article establishes triple medium seepage mathematical model “fracture—dissolution pore—matrix pore” considering lithological characteristics limestone,...

Based on the principle and characteristics of surface electromyography (sEMG) signal, this paper builds a signal acquisition system, collects signals four muscles left arm when human body performs hand movements. The EMG data were divided into 15 different sets, filtering, effective segment extraction, feature normalization, BP neural network model processing performed. Through training sets muscle combinations, comprehensive evaluation recognition accuracy complexity each model, it is found...

Abstract The Yingxiongling shale oil in the Qaidam Basin is preferred choice for Qinghai Oilfield to enter field of unconventional and gas resources, volume fracturing core technology exploration development. Through analysis geological engineering characteristics oil, it found that there are difficulties such as frequent changes vertical mechanical properties limited control range; large stress differences, making difficult form complex artificial fractures; difficulty opening multiple...

Deep learning has achieved many successes in the field of hyperspectral image (HSI) classification. Most existing deep learning-based methods have no consideration feature distribution, which may yield lowly separable and discriminative features. From perspective spatial geometry, one excellent distribution form requires to satisfy both properties, i.e., block ring. The means that a space, distance intraclass samples is close interclass far. ring represents all class are overall distributed...

Abstract Ignition and lean blowout characteristics are an important performance index of the Aeroengine combustor, which is directly related to reliability whole engine. The ignition process combustor was filmed by a CCD camera, three stages successful were obtained. By changing intake conditions, variation minimum energy in fuel-air ratio (FAR) tested. With decrease FAR, increases there sudden increase process. As room temperature pressure (LBO) FAR roughly 0.014, Increasing kerosene can...

Aiming at the problem that vibration calibration accuracy is seriously affected by additional dynamic magnetic field interference and eddy current effect of electromagnetic exciters, a suppression method based on tracking compensation proposed. The influence driving total harmonic distortion coil output force waveform characteristics are studied. set central yoke, its direction opposite to current, realize loss coil, effectively improve eliminate temperature change working platform caused...

Deep learning is a popular and effective technique for the hyperspectral image (HSI) classification. Current deep learning-based methods have numerous free parameters to be trained. They may unavailable once lacking training samples. In addition, these approaches only use features from deepest layer exclude shallow features, which kind of information loss. To remedy such deficiencies, in this work, we construct novel encoder with kernel-wise Taylor series (EKTS) HSI More specifically,...

For most hyperspectral image (HSI) classification methods, each feature code is usually individually learned, and which susceptible to noise. To remedy such deficiency, we propose a binary representation method with context -aware attention (BFCA) for HSI in this study. In model, local spectrum modules (LSMs) are first built by segmenting pixel vector into several parts computing the differences between central value its neighborhoods part. The LSMs can observe changes of spectral values,...

Hyperspectral image (HSI) classification is a widely focused topic. Existing methods export discriminative features with low-quality distribution, which may induces performance degeneration. To remedy this deficiency, we propose group-aware low-rank representation (GAL-RR) model to classify HSIs in paper. Specifically, group-wise regularization introduced capture the block-diagonal group structure of representation. With such regularization, HSI pixels from one class are well clustered into...

On account of the complex application environment and large number uncertain conditions for palletizing robot, we do path-planning multiple joints robot by algorithm based on Hierarchical Markov Decision Process. First, according to actual working environment, set range robot’s motion select conventional movement combination as basic behaviors. We can get possible reward various situations. divide state space in accordance with location information obstacle into a small clusters, sub-level...

Abstract A real gas calculation model is developed using the P-R equation of state to investigate injection and mixing characteristics supercritical RP-3 aviation kerosene in combustion chamber. In this investigation, n-decane used as an alternative fuel, physical parameters at are calculated. The jet different chamber models investigated by model. head configuration suitable for obtained analyzing spatial inhomogeneity, deviation equivalent ratio fluctuation, degree.

Given the constraints associated with a single UAV performing target tracking tasks in complex environments, this study employs multi-agent reinforcement learning approach to investigate problem of multi-directional swarm. The agent's action space and state are designed using local observable Markov model, while reward function for cluster is formulated based on potential field collision cone principle. Furthermore, address issue over-fitting during training process, Gaussian noise...

Abstract In the present study, a series of experiments on fuel jet from 3D-printed swirling nozzle were performed. The particle size distribution at different injection pressures was measured by analyzer, and enhancement atomization air also evaluated. Rosin-Rammler model used to analyze trend modulus along with inject pressure, variation analyzed development process. applicability 3D printed nozzles for aero-engine evaluated varying flow number.