Circular Convolution Research Articles

MCGnet with multi-level gated unit transformations: A time series prediction-based model for unknown pattern abnormality detection

The task of anomaly detection in data patterns remains challenging due to the diverse of fault patterns, which often render existing models ineffective when encountered with unknown or unseen data patterns in the training data. Furthermore, insufficient samples, the presence of chaotic behavior, unbalanced anomaly patterns are potential limiting factors in the practical application. While prediction error-based approaches attempt to address these challenges, the discriminability of abnormal patterns in existing scenarios is not yet sufficient, thus limiting the ability to identify abnormal phenomena. To overcome these limitations, we proposed a novel model called MCGnet. The backbone architecture of MCGnet employed a novel design of continuous multi-layer circular convolution transformations. This design enhances the expressive capability of the model by expanding the differences among data patterns. Additionally, we introduced in parallel the newly proposed temporal-frequency gated Gaussian unit, which generates richer feature representations. Subsequently, the obtained feature representations were fed into an explicit causal network to learn complex patterns of data along temporal dependencies. The MCGnet not only solves high-precision time series prediction problems but also significantly improves the discriminability between normal and abnormal patterns. Since the proposed model is accurately modeled using only a small amount of normal data, it bypasses the limiting factors such as insufficient fault samples and sample imbalance, and is able to effectively deal with various unknown data patterns. The accuracy of the proposed method was verified through simulation data (Duffing system) and its potential for engineering applications was demonstrated through real-world data (rotor system).

Efficient magnetic forward modeling for strongly magnetized bodies: An iterative integral equation approach

Magnetic surveying encounters challenges in high-precision detection and inversion on strongly magnetized bodies. Nonuniform magnetization caused by self-demagnetization and mutual magnetization makes magnetic field calculation using analytical formulas impractical. The previous discrete numerical methods have suffered from computational inefficiency or low accuracy. To resolve this issue, we develop an algorithm to solve the magnetic field integral equation for calculating the magnetization state of high-susceptibility bodies, combining the improved spatial convolution forward approach with the adaptive relaxation iteration method. To address the excessive computational burden in the magnetization field between 3D voxel arrays of the discrete model, we construct the circular convolution kernel matrix directly when the subsurface space is divided with an equidistant grid, significantly reducing redundant computations. Then, the fast Fourier transform is used to accelerate the forward discrete circular convolution operation. In addition, we derive an iterative computation scheme that adaptively adjusts the relaxation factor based on magnetic field changes to correct the effective magnetization for algorithm convergence. From a pragmatic perspective, equating the cuboid to equal-volume sphere subdivisions is developed to enhance computational efficiency further by approximately 50% despite sacrificing slight accuracy. The sphere and spherical shell models validate the algorithm accuracy, efficiency, and convergence. Furthermore, we analyze the characteristics of the mutual magnetization effect among magnetic bodies under different magnetization conditions. Finally, the applicability of the algorithm is demonstrated with a realistic example. Our algorithm shows promise for precisely exploring highly magnetized minerals, underground pipelines, and unexploded ordnance.

Nondestructive testing of runny salted egg yolk based on improved ConvNeXt-T.

Salted egg yolks from salted duck eggs are widely utilized in the domestic and international food industry as both raw materials and ingredients. When salted egg yolks are not fully cured and matured, they exist in a fluid state, with a mixture of solid and liquid internally. Due to this composition, they are susceptible to deterioration during storage and usage, necessitating their detection and classification. In this study, a dataset specifically for salted egg yolks was established, and the ConvNeXt-T model, employed as the benchmark model, underwent two notable improvements. First, a lightweight location-aware circular convolution (ParC) was introduced, utilizing a ParC-block to replace a portion of the original ConvNeXt-T block. This enhancement aimed to overcome the limitations of convolution in extracting global feature information while integrating the global sensing capability of vision transformer and the localization capability of convolution. Additionally, the activation function was modified through substitution. These improvements resulted in the final model. Experimental results indicate that the enhanced model exhibits faster convergence on the custom salted egg yolk dataset compared to the baseline model. Furthermore, a significant reduction of model parameters by a factor of 4 led to a 2.167 percentage point improvement in the accuracy of the test set. The ParC-ConvNeXt-SMU-T model achieved an accuracy of 96.833% with 26.8 million parameters. Notably, the improved model demonstrates exceptional effectiveness in recognizing salted egg yolks. PRACTICAL APPLICATION: This study can be widely applied in the process of salted egg yolk production and quality inspection, which can improve the actual sorting efficiency of salted egg yolks and reduce the labor cost at the same time. It can also be used for nondestructive testing of salted egg yolks by governmental enterprises and other regulatory authorities.

AParC-DETR: Accelerate DETR training by introducing Adaptive Position-aware Circular Convolution

AParC-DETR: Accelerate DETR training by introducing Adaptive Position-aware Circular Convolution

LSGNet: A lightweight convolutional neural network model for tomato disease identification

Tomatoes are among the most extensively grown and consumed crops worldwide, but tomato production can be greatly reduced due to various diseases. Plant diseases show different symptoms at different stages. In addition, there are similarities in the symptoms of different types of plant diseases, which hinder the recognition of diseases by existing deep learning models. Traditional convolutional neural network (CNN) models for disease recognition have a large number of parameters that require high computational resources. To overcome these challenges, we propose a lightweight CNN model named LSGNet (lightweight sandglass network) for tomato disease identification. The LSGNet backbone consists of the sandglass with efficient channel attention (SGECA) and the position aware circular convolution sandglass (ParcSG) modules. The SGECA module reduces the interference of complex environments and thus focuses on extracting useful feature information. The ParcSG module has a global receptive field, which provides more detailed feature information on disease recognition. The results show that the recognition accuracies are 92.37%, 94.32%, 89.64%, 92.70%, 94.43%, 90.97%, 89.42%, 92.98%, 89.58%, and 95.54% for AlexNet, ResNet50, VGG16, MobileNetV3-Large, ShuffleNetV2-1 × , EfficientNetV2-Small, ViT-Base, MobileViT-Small, Swin-Tiny, and LSGNet. Therefore, LSGNet has higher accuracy in recognizing tomato diseases compared to other classical models. In addition, LSGNet uses 0.75 million parameters. Compared to the lightweight CNN model MobileNetV3-Large, it only has 18% of the parameters. As a whole, the advantages of LSGNet in efficiency and lightweight structure would make it a useful tool for tomato disease recognition on mobile or embedded devices.

Solution to the issue of high-order diffraction images for cylindrical computer-generated holograms

The cylindrical computer-generated hologram (CCGH), featuring a 360° viewing zone, has garnered widespread attention. However, the issue of high-order diffraction images due to pixelated structure in CCGH has not been previously reported and solved. For a cylindrical model offering a 360° viewing zone in the horizontal direction, the high-order diffraction images always overlap with the reconstruction image, leading to quality degradation. Furthermore, the 4f system is commonly used to eliminate high-order diffraction images in planar CGH, but its implementation is predictably complex for a cylindrical model. In this paper, we propose a solution to the issue of high-order diffraction images for CCGH. We derive the cylindrical diffraction formula from the outer hologram surface to the inner object surface in the spectral domain, and based on this, we subsequently analyze the effects brought by the pixel structure and propose the high-order diffraction model. Based on the proposed high-order diffraction model, we use the gradient descent method to optimize CCGH accounting for all diffraction orders simultaneously. Furthermore, we discuss the issue of circular convolution due to the periodicity of the Fast Fourier Transform (FFT) in cylindrical diffraction. The correctness of the proposed high-order diffraction model and the effectiveness of the proposed optimization method are demonstrated by numerical simulation. To our knowledge, this is the first time that the issue of high-order diffraction images in CCGH has been proposed, and we believe our solution can offer valuable guidance to practitioners in the field.

Knowledge graph embedding in a uniform space

Knowledge graph embedding (KGE) is typically used for link prediction to automatically predict missing links in knowledge graphs. Current KGE models are mainly based on complicated mathematical associations, which are highly expressive but ignore the uniformity behind the classical bilinear translational model TransE, a model that embeds all entities of knowledge graphs in a uniform space, enabling accurate embeddings. This study analyses the uniformity of TransE and proposes a novel KGE model called ConvUs that follows uniformity with expressiveness. Based on the convolution neural network (CNN), ConvUs proposes constraints on convolution filter values and employs a multi-layer, multi-scale CNN architecture with a non-parametric L2 norm-based scoring function for the calculation of triple scores. This addresses potential uniformity-related issues in existing CNN-based KGE models, allowing ConvUs to maintain a uniform embedding space while benefiting from the powerful expressiveness of CNNs. Furthermore, circular convolution is applied to alleviate the potential orderliness contradictions, making ConvUs more suitable for conducting uniform space KGE. Our model outperformed the base model ConvKB and several baselines on the link prediction benchmark WN18RR and FB15k-237, demonstrating strong applicability and generalization and indicating that the uniformity of embedding space with high expressiveness enables more efficient knowledge graph embeddings.

Fast computation of cyclic convolutions and their applications in code-based asymmetric encryption schemes

The development of fast algorithms for key generation, encryption and decryption not only increases the efficiency of related operations. Such fast algorithms, for example, for asymmetric cryptosystems on quasi-cyclic codes, make it possible to experimentally study the dependence of decoding failure rate on code parameters for small security levels and to extrapolate these results to large values of security levels. In this article, we explore efficient cyclic convolution algorithms, specifically designed, among other things, for use in encoding and decoding algorithms for quasi-cyclic LDPC and MDPC codes. Corresponding convolutions operate on binary vectors, which can be either sparse or dense. The proposed algorithms achieve high speed by compactly storing sparse vectors, using hardware-supported XOR instructions, and replacing modulo operations with specialized loop transformations. These fast algorithms have potential applications not only in cryptography, but also in other areas where convolutions are used.

Self-supervised Learning for DNA sequences with circular dilated convolutional networks

DNA molecules commonly exhibit wide interactions between the nucleobases. Modeling the interactions is important for obtaining accurate sequence-based inference. Although many deep learning methods have recently been developed for modeling DNA sequences, they still suffer from two major issues: 1) most existing methods can handle only short DNA fragments and fail to capture long-range information; 2) current methods always require massive supervised labels, which are hard to obtain in practice. We propose a new method to address both issues. Our neural network employs circular dilated convolutions as building blocks in the backbone. As a result, our network can take long DNA sequences as input without any condensation. We also incorporate the neural network into a self-supervised learning framework to capture inherent information in DNA without expensive supervised labeling. We have tested our model in two DNA inference tasks, the human variant effect and the open chromatin region of plants, where the experimental results show that our method outperforms five other deep learning models. Our code is available at https://github.com/wiedersehne/cdilDNA.

The Bernoulli clock: probabilistic and combinatorial interpretations of the Bernoulli polynomials by circular convolution

AbstractThe factorially normalized Bernoulli polynomials $b_n(x) = B_n(x)/n!$ are known to be characterized by $b_0(x) = 1$ and $b_n(x)$ for $n \gt 0$ is the anti-derivative of $b_{n-1}(x)$ subject to $\int _0^1 b_n(x) dx = 0$ . We offer a related characterization: $b_1(x) = x - 1/2$ and $({-}1)^{n-1} b_n(x)$ for $n \gt 0$ is the $n$ -fold circular convolution of $b_1(x)$ with itself. Equivalently, $1 - 2^n b_n(x)$ is the probability density at $x \in (0,1)$ of the fractional part of a sum of $n$ independent random variables, each with the beta $(1,2)$ probability density $2(1-x)$ at $x \in (0,1)$ . This result has a novel combinatorial analog, the Bernoulli clock: mark the hours of a $2 n$ hour clock by a uniformly random permutation of the multiset $\{1,1, 2,2, \ldots, n,n\}$ , meaning pick two different hours uniformly at random from the $2 n$ hours and mark them $1$ , then pick two different hours uniformly at random from the remaining $2 n - 2$ hours and mark them $2$ , and so on. Starting from hour $0 = 2n$ , move clockwise to the first hour marked $1$ , continue clockwise to the first hour marked $2$ , and so on, continuing clockwise around the Bernoulli clock until the first of the two hours marked $n$ is encountered, at a random hour $I_n$ between $1$ and $2n$ . We show that for each positive integer $n$ , the event $( I_n = 1)$ has probability $(1 - 2^n b_n(0))/(2n)$ , where $n! b_n(0) = B_n(0)$ is the $n$ th Bernoulli number. For $ 1 \le k \le 2 n$ , the difference $\delta _n(k)\,:\!=\, 1/(2n) -{\mathbb{P}}( I_n = k)$ is a polynomial function of $k$ with the surprising symmetry $\delta _n( 2 n + 1 - k) = ({-}1)^n \delta _n(k)$ , which is a combinatorial analog of the well-known symmetry of Bernoulli polynomials $b_n(1-x) = ({-}1)^n b_n(x)$ .

ANEW ARCHITECTURE OF KEY GENERATION USING DWT FOR IMAGE ENCRYPTION WITH THREE LEVELS ARNOLD TRANSFORM PERMUTATION

The security of image transmission is an important issue in digital communication. As well as it is necessary to preserve and protect important information for several applications like military, medical and other services related to high confidentiality over the Internet or other unprotected networks. In this paper a proposed encryption scheme was introduced that using Lorenz system with a circular convolution and discreet cosine transform (DCT). The diffusion process was achieved using three levels of Arnold transform permutation: - namely, block level, inside each block and pixel level. The image was divided into blocks of sizes 8x8 pixels and shuffle by applying Fisher-Yates permutation image pixels. The DCT was applied to each block and multiplied by the H kernel matrix to achieve the circular convolution. Next the logistic map is used for diffusion to get the cipher image. A new key generation method is applied in order to produce the key values for generating the chaos numbers sequence. Finally applying the discreet wavelet transform (DWT) to the image will produced four quarters (LL, LH, HL and HH). The cosine of each pixel in the LL and HH quarters were computed. The sine of each pixel in the LH and HL quarters were also computed in order to generate the keys for the chaos sequence. By using this way of keys generation in every image, the differential attack possibility will be negligible

Series Arc Fault Detection Method Based on Signal-Type Enumeration and Zoom Circular Convolution Algorithm

AC series arc fault (SAF) detection under various circuits is always a challenging task because arcing current is affected by different circuit types and fault points. Additionally, normal current is confused as arcing current when non-linear loads are present in the circuit. To cope with these issues, this paper presents a detection method using signal-type enumeration and a zoom circular convolution (CC) (ZCC) algorithm. The generalization ability of the detection method under unknown conditions is improved by means of signal-type enumeration. The impulsive components of current can be divided into stable, periodically impulsive, non-periodically impulsive and hybrid signals by enumeration. Then, given the CC limitations, the ZCC is presented to extract distinguishable features and decrease the computation complexity of CC, and the signal function is partially reconstructed to improve the CC performance. Finally, the presented detection method is analyzed in a laptop and evaluated by TMS320F28335. The online detection results show that the proposed method determined by known single-load circuits has good detection accuracy under unknown conditions and can be achieved in practical application.

EEG-CDILNet: a lightweight and accurate CNN network using circular dilated convolution for motor imagery classification.

Objective.The combination of the motor imagery (MI) electroencephalography (EEG) signals and deep learning-based methods is an effective way to improve MI classification accuracy. However, deep learning-based methods often need too many trainable parameters. As a result, the trade-off between the network decoding performance and computational cost has always been an important challenge in the MI classification research.Approach.In the present study, we proposed a new end-to-end convolutional neural network (CNN) model called the EEG-circular dilated convolution (CDIL) network, which takes into account both the lightweight model and the classification accuracy. Specifically, the depth-separable convolution was used to reduce the number of network parameters and extract the temporal and spatial features from the EEG signals. CDIL was used to extract the time-varying deep features that were generated in the previous stage. Finally, we combined the features extracted from the two stages and used the global average pooling to further reduce the number of parameters, in order to achieve an accurate MI classification. The performance of the proposed model was verified using three publicly available datasets.Main results.The proposed model achieved an average classification accuracy of 79.63% and 94.53% for the BCIIV2a and HGD four-classification task, respectively, and 87.82% for the BCIIV2b two-classification task. In particular, by comparing the number of parameters, computation and classification accuracy with other lightweight models, it was confirmed that the proposed model achieved a better balance between the decoding performance and computational cost. Furthermore, the structural feasibility of the proposed model was confirmed by ablation experiments and feature visualization.Significance.The results indicated that the proposed CNN model presented high classification accuracy with less computing resources, and can be applied in the MI classification research.

Global and pyramid convolutional neural network with hybrid attention mechanism for hyperspectral image classification

Convolutional neural networks (CNNs) have shown impressive results in the hyperspectral image (HSI) classification. However, they still face certain limitations that can impact their effectiveness. The kernels in standard convolution have fixed spatial sizes and spectral depths, which cannot capture global semantic features from HSI and the existing methods for extracting multiscale information are inadequate. Based on these, we propose a global and pyramid convolutional network with a hybrid attention mechanism (GPHANet) for HSI classification. GPHANet adopts a two-branch architecture to extract both local and global hyperspectral features, leveraging the strengths of each to enhance classification performance. The local branch employs dynamic pyramid convolution, which customizes parameters for convolutional kernels of multiple scales based on the input image. This design enables the model to accurately and comprehensively extract multiscale feature information, capturing subtle variations in the HSI. The global branch utilizes circular convolution with a global receptive field, enabling each position in the convolutional layers to gather information from the entire input space. This allows the model to learn global contextual features, enhancing its understanding of the overall structure and semantics of the HSI. After extracting deep global and local features, we apply a compact hybrid attention mechanism (HAM) to capture long-range dependencies along both spatial and spectral dimensions, resulting in a more discriminative feature representation. On the Pavia University, Salinas, and Hong Hu datasets, the proposed model achieved overall accuracies of 99.22%, 98.74%, and 96.80%, respectively, using limited training samples. These results are much better compared to the state-of-the-art methods.

W-Net: A boundary-enhanced segmentation network for stroke lesions

Accurate lesion segmentation is a critical technology basis for the treatment and prognosis of stroke. Stroke lesion segmentation suffers from complex background and noise interferes due to the similar grayscale characteristics of tissue structures and imaging mode in medical image. To resolve the difficulty of fuzzy boundary for lesion segmentation, we propose a novel two-stage brain MRI lesion segmentation method, called W-Net. W-Net leverages the CNN and transformer-based method as the backbone network, and introduces boundary deformation module (BDM), boundary constraint module (BCM) to handle the tough condition of fuzzy boundary. The BDM utilizes circular convolution to correct the initial boundary and BCM exploits dilated convolution to dynamically constrain the object boundary. Moreover, we design a multi-task learning loss function to optimize the W-Net from both region and boundary perspectives. Extensive experiments are conducted using the two ischemic stroke lesion segmentation datasets, and the results are compared with other advanced models. The experimental results demonstrate that the proposed method is effective in preserving the edge information of stroke lesions, thus achieving competitive performance on the segmentation task.

Hybrid kalman filtering algorithm with wavelet packet data processing for linear dynamical systems

The paper develops a hybrid algorithm for predicting a linear dynamic system based on a combination of an adaptive Kalman filter with preprocessing using a wavelet packet analysis of the initial data of the background of the system under study.
 Being based on Fourier analysis, wavelet analysis and wavelet packet analysis are quite acceptable for time-frequency analysis of a signal, but they cannot be performed recursively and in real time and, therefore, cannot be used for dynamic analysis of random processes. In combination with the Kalman filter, a combination of the characteristics of the multiple-resolution wavelet transform and the recurrent formulas of the Kalman filter in real time is obtained.
 Since the original signal is usually given in the form of discrete measurements, to implement their convolution used in the Kalman filter, it is necessary to use cyclic convolutions with periodic continuation of the signal for any time interval. In the case of different values of the original signal at the ends of the considered time interval [0,T], the periodized signal can have large values and sharp different amplitude at the ends of the periodization interval.
 To smooth out the values of the periodized signal at the ends of the periodization interval, a cascade decomposition and recovery algorithm was used using Dobshy boundary wavelets with a finite number of moments. Signal recovery is performed in a series of operations comparable to the duration of the time interval under consideration.
 The smoothed signal obtained in this way is used as a Kalman filter platform for predicting the dynamic system under study.
 Taking into account that the correlation functions of the noise in the observation equation and the phase state of the system are usually unknown, the adaptation of the Kalman filter to these noises (interference) is carried out on the basis of a zeroing sequence. The manuscript does not contain related data

Variable Binding for Sparse Distributed Representations: Theory and Applications.

Variable binding is a cornerstone of symbolic reasoning and cognition. But how binding can be implemented in connectionist models has puzzled neuroscientists, cognitive psychologists, and neural network researchers for many decades. One type of connectionist model that naturally includes a binding operation is vector symbolic architectures (VSAs). In contrast to other proposals for variable binding, the binding operation in VSAs is dimensionality-preserving, which enables representing complex hierarchical data structures, such as trees, while avoiding a combinatoric expansion of dimensionality. Classical VSAs encode symbols by dense randomized vectors, in which information is distributed throughout the entire neuron population. By contrast, in the brain, features are encoded more locally, by the activity of single neurons or small groups of neurons, often forming sparse vectors of neural activation. Following Laiho et al. (2015), we explore symbolic reasoning with a special case of sparse distributed representations. Using techniques from compressed sensing, we first show that variable binding in classical VSAs is mathematically equivalent to tensor product binding between sparse feature vectors, another well-known binding operation which increases dimensionality. This theoretical result motivates us to study two dimensionality-preserving binding methods that include a reduction of the tensor matrix into a single sparse vector. One binding method for general sparse vectors uses random projections, the other, block-local circular convolution, is defined for sparse vectors with block structure, sparse block-codes. Our experiments reveal that block-local circular convolution binding has ideal properties, whereas random projection based binding also works, but is lossy. We demonstrate in example applications that a VSA with block-local circular convolution and sparse block-codes reaches similar performance as classical VSAs. Finally, we discuss our results in the context of neuroscience and neural networks.

Real-time 3D human pose estimation without skeletal a priori structures

This study is about real-time 2D-3D human pose estimation without using the a priori structure of the skeleton and with a low number of parameters for regression tasks. Current graph convolution-based 3D human pose tasks require structural knowledge of the skeleton, which limits the exploration of pose estimation for unknown structures. Inspired by tyre rotation and circular convolution, weights rotation is used to fully learn the potential connections between the human joints. We refer to this process as the weight cyclic sharing mechanism, a novel method for updating features. It does not require knowledge of the structure of the human skeleton and learns different constraints between joints with a low number of parameters. We propose an end-to-end weight circular sharing network (WCirSNet) based on the weight circular sharing mechanism. We propose a simple and efficient weighted residual block in this WCirSNet. The superiorities of the weight circular sharing mechanism and weighted residual block were verified by abundant ablation studies. Extensive evaluations on two challenging benchmark datasets (Human 3.6 M, MPI-INF-3DHP) show that the performance and generalization capabilities of our framework are superior to the results of many previously advanced methods.

A Scoping Review of Cerebral Doppler Arterial Waveforms in Infants.

Cerebral Doppler ultrasound has been an important tool in pediatric diagnostics and prognostics for decades. Although the Doppler spectrum can provide detailed information on cerebral perfusion, the measured spectrum is often reduced to simple numerical parameters. To help pediatric clinicians recognize the visual characteristics of disease-associated Doppler spectra and identify possible areas for future research, a scoping review of primary studies on cerebral Doppler arterial waveforms in infants was performed. A systematic search in three online bibliographic databases yielded 4898 unique records. Among these, 179 studies included cerebral Doppler spectra for at least five infants below 1 y of age. The studies describe variations in the cerebral waveforms related to physiological changes (43%), pathology (62%) and medical interventions (40%). Characteristics were typically reported as resistance index (64%), peak systolic velocity (43%) or end-diastolic velocity (39%). Most studies focused on the anterior (59%) and middle (42%) cerebral arteries. Our review highlights the need for a more standardized terminology to describe cerebral velocity waveforms and for precise definitions of Doppler parameters. We provide a list of reporting variables that may facilitate unambiguous reports. Future studies may gain from combining multiple Doppler parameters to use more of the information encoded in the Doppler spectrum, investigating the full spectrum itself and using the possibilities for long-term monitoring with Doppler ultrasound.

A hybrid neural architecture search for hyperspectral image classification

Convolution neural network (CNN)is widely used in hyperspectral image (HSI) classification. However, the network architecture of CNNs is usually designed manually, which requires careful fine-tuning. Recently, many technologies for neural architecture search (NAS) have been proposed to automatically design networks, further improving the accuracy of HSI classification to a new level. This paper proposes a circular kernel convolution-β-decay regulation NAS-confident learning rate (CK-βNAS-CLR) framework to automatically design the neural network structure for HSI classification. First, this paper constructs a hybrid search space with 12 kinds of operation, which considers the difference between enhanced circular kernel convolution and square kernel convolution in feature acquisition, so as to improve the sensitivity of the network to hyperspectral information features. Then, the β-decay regulation scheme is introduced to enhance the robustness of differential architecture search (DARTS) and reduce the discretization differences in architecture search. Finally, we combined the confidence learning rate strategy to alleviate the problem of performance collapse. The experimental results on public HSI datasets (Indian Pines, Pavia University) show that the proposed NAS method achieves impressive classification performance and effectively improves classification accuracy.