Uniform Sampling Research Articles

Optimized uniform sampling and validation of fiber tracts from magnetic resonance tractography for in vivo architectural measurement of human forearm muscles.

In vivo muscle architectural parameters can be calculated from the fiber tracts using magnetic resonance (MR) tractography. However, the reconstructed tracts may be unevenly distributed within the muscle volume and there lacks commonly used metric to quantitatively evaluate the validity of the tracts. Our objective is to measure forearm muscle architecture by uniformly sampling fiber tracts from the candidate streamlines in MR tractography and validate the reconstructed fiber tracts qualitatively and quantitatively. We proposed farthest streamline sampling (FSS) to uniformly sample fiber tracts from the candidate streamlines. The method was evaluated on the MR data acquired from 12 healthy subjects for 17 forearm muscles and was compared with two conventional methods through uniform coverage performance. Anatomical correctness was verified by: 1. visually assessing fiber orientation, 2. checking whether architectural parameters were within physiological ranges and 3. classifying architectural types. The proposed FSS yielded optimal uniform coverage performance among the three methods (P<0.05). FSS reduced the sampling of long tracts (10% fiber length reduction, P<0.05), and the estimated architectural parameters were within the physiological ranges (P<0.05). The tractography visually matched cadaveric specimens. The architectural types of 16 muscles were correctly classified except for the palmaris longus, which exhibited a linear arrangement of fiber endpoints (R2 = 0.95±0.02, P<0.001). The proposed FSS method reconstructed uniformly distributed fiber tracts and the anatomical correctness of the reconstructed tracts was verified. The novel methods allow for accurate in vivo muscle architectural measurement, which was demonstrated through the characterization of architectural properties in human forearm muscles.

Computer-Vision-Oriented Adaptive Sampling in Compressive Sensing.

Compressive sensing (CS) is recognized for its adeptness at compressing signals, making it a pivotal technology in the context of sensor data acquisition. With the proliferation of image data in Internet of Things (IoT) systems, CS is expected to reduce the transmission cost of signals captured by various sensor devices. However, the quality of CS-reconstructed signals inevitably degrades as the sampling rate decreases, which poses a challenge in terms of the inference accuracy in downstream computer vision (CV) tasks. This limitation imposes an obstacle to the real-world application of existing CS techniques, especially for reducing transmission costs in sensor-rich environments. In response to this challenge, this paper contributes a CV-oriented adaptive CS framework based on saliency detection to the field of sensing technology that enables sensor systems to intelligently prioritize and transmit the most relevant data. Unlike existing CS techniques, the proposal prioritizes the accuracy of reconstructed images for CV purposes, not only for visual quality. The primary objective of this proposal is to enhance the preservation of information critical for CV tasks while optimizing the utilization of sensor data. This work conducts experiments on various realistic scenario datasets collected by real sensor devices. Experimental results demonstrate superior performance compared to existing CS sampling techniques across the STL10, Intel, and Imagenette datasets for classification and KITTI for object detection. Compared with the baseline uniform sampling technique, the average classification accuracy shows a maximum improvement of 26.23%, 11.69%, and 18.25%, respectively, at specific sampling rates. In addition, even at very low sampling rates, the proposal is demonstrated to be robust in terms of classification and detection as compared to state-of-the-art CS techniques. This ensures essential information for CV tasks is retained, improving the efficacy of sensor-based data acquisition systems.

Poisson tensor completion with transformed correlated total variation regularization

Tensor completion involves recovering the underlying tensor from partial observations, and in this paper we focus on the point that these observations obey the Poisson distribution. To contend with this problem, we adopt a popular method that minimizes the sum of the data-fitting term and the regularization term under a uniform sampling mechanism. Specifically, we consider the negative logarithmic maximum likelihood estimate of the Poisson distribution as the data-fitting term. To effectively characterize the intrinsic structure of the tensor data, we propose a parameter-free regularization term that can simultaneously capture the low rankness and local smoothness of the underlying tensor. Here, the transformed tensor nuclear norm is used to explore the low rankness under suitable unitary transformations. We present theoretical derivations to demonstrate the feasibility of the proposed model. Furthermore, we develop an algorithm based on the alternating direction multiplier method (ADMM) to efficiently solve the proposed optimization problem, with its overall convergence being established. A series of numerical experiments show that proposed model yields a pleasing accuracy over several state-of-the-art models.

A Security Domain Feature Pattern Recognition Method Based on Capsule Neural Network

Abstract In this paper, a security domain feature pattern recognition method based on capsule neural network is proposed. Firstly, Halton sampling is performed on the active output of all generators except for the balancing machine to obtain a uniform sample in the coverage space. Then, the support vector machine based recursive feature elimination (SVM-RFE) method is used to select the critical pre fault feature quantities of the system. Finally, based on the selected active features, one-dimensional data is converted into two-dimensional data and input into the capsule neural network. Dynamic security domain boundaries are constructed in both two-dimensional and three-dimensional spaces, which can divide the feature space into stable and unstable subspaces.

Systematically Revisiting All NuSTAR Spins of Black Holes in X-Ray Binaries

We extend our recent work on black hole spin in X-ray binary systems to include an analysis of 189 archival NuSTAR observations from 24 sources. Using self-consistent data reduction pipelines, spectral models, and statistical techniques, we report an unprecedented and uniform sample of 36 stellar-mass black hole spin measurements based on relativistic reflection. This treatment suggests that prior reports of low spins in a small number of sources were generally erroneous: our comprehensive treatment finds that those sources tend to harbor black holes with high spin values. Overall, within 1σ uncertainty, ∼86% of the sample are consistent with a ≥ 0.95, ∼94% of the sample are consistent with a ≥ 0.9, and 100% are consistent with a ≥ 0.7 (the theoretical maximum for neutron stars; a = cJ/GM 2). We also find that the high-mass X-ray binaries (those with A-, B-, or O-type companions) are consistent with a ≥ 0.9 within the 1σ errors; this is in agreement with the low-mass X-ray binary population and may be especially important for comparisons to black holes discovered in gravitational wave events. In some cases, different spectra from the same source yield similar spin measurements but conflicting values for the inclination of the inner disk; we suggest that this is due to variable disk winds obscuring the blue wing of the relativistic Fe K emission line. We discuss the implications of our measurements, the unique view of systematic uncertainties enabled by our treatment, and future efforts to characterize black hole spins with new missions.

Reduced effectiveness of permethrin-treated military uniforms after prolonged wear measured by contact irritancy and toxicity bioassays with Ixodes scapularis (Acari: Ixodidae) nymphs.

Factory-treated permethrin uniforms are the primary method used by the US Army to prevent arthropod bites and transmission of arthropod-borne diseases. In this study previously worn uniforms were collected from cadets at the United States Military Academy in West Point, NY to determine the amount of permethrin remaining after prolonged wear and subsequent effects on ticks. Six trousers were collected from cadets in the sophomore, junior, and senior classes. A new, unwashed uniform served as a positive control and an untreated maternity uniform served as a negative control. Fabric samples were removed from each trouser and used in bioassays with laboratory-reared Ixodes scapularis Say nymphs. Contact irritancy bioassays measured the nymphs' ability to remain in contact with fabric for a period of 5 min. Toxicity bioassays measured tick survival at 1 and 24 h after contacting uniform samples. liquid chromatography-mass spectrometry was used to quantify the permethrin content (mg/cm2) in each uniform after the bioassays were complete. The results showed significant amounts of permethrin were lost after extended wear and it was related to the cadet year group. The contact irritancy assays found uniforms with less permethrin did not irritate ticks and cause them to dislodge. Mortality was also affected by permethrin levels, with less ticks dying at 24 h on older uniforms. The results from this study show older uniforms lose most of their permethrin and no longer provide the same levels of protection.

A Novel Fast Iterative STAP Method with a Coprime Sampling Structure.

In space-time adaptive processing (STAP), the coprime sampling structure can obtain better clutter suppression capabilities at a lower hardware cost than the uniform linear sampling structure. However, in practical applications, the performance of the algorithm is often limited by the number of training samples. To solve this problem, this paper proposes a fast iterative coprime STAP algorithm based on truncated kernel norm minimization (TKNM). This method establishes a virtual clutter covariance matrix (CCM), introduces truncated kernel norm regularization technology to ensure the low rank of the CCM, and transforms the non-convex problem into a convex optimization problem. Finally, a fast iterative solution method based on the alternating direction method is presented. The effectiveness and accuracy of the proposed algorithm are verified through simulation experiments.

Design and Multi-Objective Optimization for Improving Torque Performance of a Permanent Magnet-Assisted Synchronous Reluctance Motor

Permanent magnet-assisted synchronous reluctance motors (PMA-SynRMs) are widely used in various industries as a relatively inexpensive and high-performance energy conversion device. The model proposed in this article relies on a magnetic pole-biased permanent magnet synchronous reluctance motor with a magnetic focusing effect. Two types of models with Halbach array and magnetic focusing effect have been proposed, which increase excitation and make the internal magnetic circuit of the rotor more saturated, thereby achieving higher electromagnetic torque. Through finite element simulation analysis and verification, the motor characteristics of the basic and proposed permanent magnet-assisted synchronous reluctance motor were calculated, including the air gap flux density and back electromotive force (EMF) in no-load analysis, as well as the average torque, torque ripple, and efficiency in load analysis. In addition, multi-objective optimization was also conducted on the rotor topology structure of proposed model two, using the uniform Latin hypercube sampling method to uniformly sample the data samples and the Pearson correlation coefficients to perform a sensitivity analysis on the data. The pilOPT multi-objective autonomous optimization algorithm was used to perform multi-objective autonomous optimization on parameters with high correlation, and the best-found solution based on the Pareto front was selected. Compared with proposed model two, the average torque of the optimized model increased by 18.14%, the efficiency increased by 1.05% and the torque ripple decreased by 5.22%. Finally, the anti-demagnetization performance of the optimized model’s permanent magnet was analyzed.

Uniform-sampling foveated Fourier single-pixel imaging

Inspired by the retina of human eye, foveated imaging is an effective way to achieve high imaging quality and high imaging efficiency. However, with an inherent property of nonuniformity, advantages of foveated imaging cannot be played well due to the confliction with many theories having uniform properties. Thus, an innovative concept “uniform-sampling foveated imaging” is proposed to break the barrier of foveated imaging and uniform imaging. In this work, the concept is used as a uniform-sampling foveated Fourier single-pixel imaging (UFFSI) to achieve high imaging quality and high imaging efficiency of single-pixel imaging (SPI). First, by flexibly using the proposed three kinds of foveated pattern structures of foveated SPI, namely, “circular structure” via log-polar transformation, “rectangular structure” via log-rectilinear transformation and “rotating-rectangle structure” via log-rectilinear-rotation transformation, the total data number is reduced radically by the data redundancy reduction only requiring high resolution (HR) on regions of interest (ROIs). Next, by a nonuniform weight distribution processing based on the innovative concept of “uniform-sampling foveated imaging”, nonuniform sampling is transformed into uniform sampling to obtain the same uniform optical power as the uniform HR FSI, thus, sparse spectrum property and fast Fourier transform of FSI are used accurately and directly to obtain high imaging quality with further reduced measurements and fast reconstruction. Experimentally, in a large-scale scene, at an ultralow sampling ratio of 0.78 % referring to uniform HR FSI with 1024 × 768 pixels, by using UFFSI with 89 % reduction of data redundancy, ROI has a significantly better imaging quality with around 3.3times reduction in reconstruction time. We hope this work can provide a breakthrough point for future foveated imaging and real-time SPI in real life.

Imbalanced generative sampling of training data for improving quality of machine learning model

Design exploration in engineering applications often requires a meticulous experimental or numerical study to evaluate performance (Y) of each design, which may require great effort, time or resources. Reducing the number of these tests for finding a good design is of paramount importance in all engineering fields. This study aims at computing a machine learning (ML) model using less number of designs as training data. Uniform sampling (US) in the design space (based on predefined design parameters) to obtain a training data is a promising approach. We further extend this sampling concept to obtain designs in the design space by also employing the ML model.The designs are selected via two non-uniform (imbalanced) sampling methods (namely, height-based sampling - HBS and gradient-based sampling - GBS) while considering their Y and gradient, dY, values. These values are divided into uniform intervals, and we aim at equalizing the number of designs in the training data at each interval as much as possible. This can force designs to have minimum or maximum Y or dY values, which, in fact, lie on small portion of the design space, in general. Therefore, capturing designs from all design space portions can be enabled.Results of the proposed methods are compared against US along with two well studied non-uniform sampling strategies, Stratified Over Sampling (SOS) and Gaussian-Process Based Sampling (GPBS). To reliably investigate quality of ML models obtained using designs sampled via US, SOS, GPBS, HBS and GBS, we utilize standard test (known) functions (such as Easom and Beale) as substitutes for engineering problems. According to the results presented, ML models using HBS and GBS have either better prediction accuracy or wider applicability compared to all other tested sampling methods.

Faster Counting and Sampling Algorithms using Colorful Decision Oracle

In this work, we consider d - Hyperedge Estimation and d - Hyperedge Sample problems that deal with estimation and uniform sampling of hyperedges in a hypergraph ℋ( U (ℋ), ℱ(ℋ) in the query complexity framework, where U (ℋ) denotes the set of vertices and ℱ(ℋ) denotes the set of hyperedges. The oracle access to the hypergraph is called Colorful Independence Oracle ( CID ), which takes d (non-empty) pairwise disjoint subsets of vertices A 1 ,..., A d ⊆ U (ℋ) as input and answers whether there exists a hyperedge in ℋ having exactly one vertex in each A i for all i ∈ {1, 2, ..., d }. Apart from the fact that d - Hyperedge Estimation and d - Hyperedge Sample problems with CID oracle access seem to be nice combinatorial problems, Dell et al. [SODA’20 &amp; SICOMP’22] established that decision vs. counting complexities of a number of combinatorial optimization problems can be abstracted out as d - Hyperedge Estimation problem with a CID oracle access. The main technical contribution of this article is an algorithm that estimates m = |ℱ(ℋ)| with \(\widehat{m}\) such that \(\begin{equation*} \frac{1}{C_{d}\log ^{d-1} n} \;\le \; \frac{\widehat{m}}{m} \;\le \; C_{d} \log ^{d-1} n \end{equation*}\) by using at most C d log d +2 n CID queries, where n denotes the number of vertices in the hypergraph ℋ and C d is a constant that depends only on d . Our result, when coupled with the framework proposed by Dell et al. (SODA’20 &amp; SICOMP’22), leads to implies improved bounds for (1 ± ε)-approximation (where ε ∈ (0,1)) for the following fundamental problems: Edge Estimation using the Bipartite Independent Set ( BIS ) query. We improve the bound obtained by Beame et al. (ITCS’18 &amp; TALG’20). Triangle Estimation using the Tripartite Independent Set ( TIS ) query. Currently, Dell et al.’s result gives the best bound for the case of triangle estimation in general graphs (SODA’20 &amp; SICOMP’22). The previous best bound for the case of graphs with low co-degree (co-degree of a graph is the maximum number of triangles incident over any edge of the graph) was due to Bhattacharya et al. (ISAAC’19 &amp; TOCS’21). We improve both of these bounds. Hyperedge Estimation &amp; Sampling using Colorful Independence Oracle ( CID ). We give an improvement over the bounds obtained by Dell et al. (SODA’20 &amp; SICOMP’22).

3D spatial distribution of soil pollutants based on geo-shadowing anisotropic RBF-PCA

Research on soil contamination has become increasingly important, but there is limited information about where to sample for pollutants. Thus, the use of three-dimensional (3D) spatial interpolation techniques has been promoted in this area of study. However, the application of traditional interpolation methods is limited in geography, especially in the expression of anisotropy, and it is not associated with geographical properties. To address this issue, we used a test site (a factory in Nanjing) to develop a new research method based on the geographical shading radial basis function (RBF) interpolation method, which considers 3D anisotropy and geographical attribute expression. Drilling and uniform sampling were used to sample the contaminated area at this test site. This approach included two steps: i) An ellipsoid with anisotropic properties was constructed. Thus, the first step was to determine the shape of the ellipsoid using principal component analysis (PCA) to determine the main orientations and construct a rotational and stretched matrix. The second step was determining the ellipsoid size by computing the range using the variogram method for orientations. ii) During field measurement, the geospatial direction influences soil attribute values, so a shadowing calculation method was derived for quadratic weight determination. Then, the weight of the attribute value of known points can be assigned to meet the field conditions. Lastly, the model was evaluated using the root mean square error (RMSE). For the 2D space, the RMSE values of Kriging, RBF, and the proposed method are 6.09, 7.12, and 5.02, respectively. The R2 values of Kriging, RBF, and the proposed method are 0.871, 0.832, and 0.946, respectively. For the 3D space, the RMSE values of Kriging, RBF, and the proposed method are 2.65, 2.23, and 2.58, respectively. The R2 values of Kriging, RBF, and the proposed method are 0.934, 0.912, and 0.953, respectively. The resulting fitted model was relatively smooth and met experimental needs. Thus, we believe that the interpolation method can be applied as a new method to predict the distribution of soil pollutants.

Reservoir Sampling over Joins

Sampling over joins is a fundamental task in large-scale data analytics. Instead of computing the full join results, which could be massive, a uniform sample of the join results would suffice for many purposes, such as answering analytical queries or training machine learning models. In this paper, we study the problem of how to maintain a random sample over joins while the tuples are streaming in. Without the join, this problem can be solved by some simple and classical reservoir sampling algorithms. However, the join operator makes the problem significantly harder, as the join size can be polynomially larger than the input. We present a new algorithm for this problem that achieves a near-linear complexity. The key technical components are a generalized reservoir sampling algorithm that supports a predicate, and a dynamic index for sampling over joins. We also conduct extensive experiments on both graph and relational data over various join queries, and the experimental results demonstrate significant performance improvement over the state of the art.

Модели составных гармонических полуволн и связь дискретизации времени с энтропией временных параметров сигналов

The problem of finding the absolute error of stepwise and linear interpolation of the control signal from uniform samples from it using models of composite harmonic half-waves is solved. Previously, during the inspection of the control object, the maximum values of the signal parameters and half-waves are determined: speed, acceleration and sharpness, there are no spectrum parameters. To determine the values of the intervals of uniform sampling of time, two groups of models of "harmonic half-waves" are considered. The first group of models is described by harmonic time functions whose parameters are consistent. The second group of models is described by composite harmonic functions of time, thereby the time parameters of the signals are consistent. It is proved that with an increase in the entropy of the maximum values of the signal parameters, the value of the time sampling interval increases without increasing the interpolation error. Thus, the entropy value of the signal parameters serves as an indicator of their inconsistency. The results of modeling and graphs obtained in the environment of the mathematical package are presented. The results are intended to optimize the loading of input tasks and primary information processing of processors in robust real-time automation systems, for example, used to control high-speed trains when braking in sliding or skidding mode.

Recovery of Discrete-Time Signal Based on the Moving Average Model and Estimation of the Samples Correlation in Forward and Reverse Forecasting

The article discusses the development of mathematical support for the recovery of the values of discrete-time sequence samples obtained as a result of uniform sampling of a continuous signal. The recovery problem of discrete-time sequence samples is solved for a signal that can be considered stationary or stationary at least in a broad sense (quasi-stationary). The development of mathematical support for the recovery of the values of signal samples was carried out on the basis of constructing a moving average model and estimating the correlation of signal samples over time with forward and reverse forecasting. Estimates of the signal correlation function necessary to recover sample sections with lost values are calculated from samples with known values. Correlation function estimates can be calculated regardless of the location of the recovery area when the condition of stationarity of the signal is met. The obtained estimates of the correlation function samples can be used for both forward and reverse forecasting. Moreover, even if it is necessary to recover several problem sections, it is enough to calculate only once the sample of correlation function estimates necessary for their restoration. The resulting mathematical solution to the problem became the basis for the development of algorithmic support. Test tests and functional checks of the algorithmic support were carried out on the basis of simulation using a signal model representing an additive sum of harmonic components with random initial phases. The simulation results showed that the calculation of estimates of the lost sample values is carried out with a fairly low error, both in forward and reverse forecasting, as well as when they are used together. In practice, the choice of a sequence recovery algorithm based on forward or reverse forecasting will be determined based on the actual conditions of its processing. In particular, if previous samples with known values are not enough to carry out forward forecasting, then the reverse forecasting procedure is implemented and vice versa. The developed algorithmic support can be implemented in the form of metrologically significant software for digital signal processing systems.

Quantum-Inspired Classical Algorithm for Graph Problems by Gaussian Boson Sampling

We present a quantum-inspired classical algorithm that can be used for graph-theoretical problems, such as finding the densest k subgraph and finding the maximum weight clique, which are proposed as applications of a Gaussian boson sampler. The main observation from Gaussian boson samplers is that a given graph’s adjacency matrix to be encoded in a Gaussian boson sampler is non-negative and that computing the output probability of Gaussian boson sampling restricted to a non-negative adjacency matrix is thought to be strictly easier than general cases. We first provide how to program a given graph problem into our efficient classical algorithm. We then numerically compare the performance of ideal and lossy Gaussian boson samplers, our quantum-inspired classical sampler, and the uniform sampler for finding the densest k subgraph and finding the maximum weight clique and show that the advantage from Gaussian boson samplers is not significant in general. We finally discuss the potential advantage of a Gaussian boson sampler over the proposed quantum-inspired classical sampler. Published by the American Physical Society 2024

A uniform sampling method for permutation space

A uniform sampling method for permutation space

AARF: Autonomous Attack Response Framework for Honeypots to Enhance Interaction Based on Multi-Agent Dynamic Game

Highly interactive honeypots can form reliable connections by responding to attackers to delay and capture intranet attacks. However, current research focuses on modeling the attacker as part of the environment and defining single-step attack actions by simulation to study the interaction of honeypots. It ignores the iterative nature of the attack and defense game, which is inconsistent with the correlative and sequential nature of actions in real attacks. These limitations lead to insufficient interaction of the honeypot response strategies generated by the study, making it difficult to support effective and continuous games with attack behaviors. In this paper, we propose an autonomous attack response framework (named AARF) to enhance interaction based on multi-agent dynamic games. AARF consists of three parts: a virtual honeynet environment, attack agents, and defense agents. Attack agents are modeled to generate multi-step attack chains based on a Hidden Markov Model (HMM) combined with the generic threat framework ATT&amp;CK (Adversarial Tactics, Techniques, and Common Knowledge). The defense agents iteratively interact with the attack behavior chain based on reinforcement learning (RL) to learn to generate honeypot optimal response strategies. Aiming at the sample utilization inefficiency problem of random uniform sampling widely used in RL, we propose the dynamic value label sampling (DVLS) method in the dynamic environment. DVLS can effectively improve the sample utilization during the experience replay phase and thus improve the learning efficiency of honeypot agents under the RL framework. We further couple it with a classic DQN to replace the traditional random uniform sampling method. Based on AARF, we instantiate different functional honeypot models for deception in intranet scenarios. In the simulation environment, honeypots collaboratively respond to multi-step intranet attack chains to defend against these attacks, which demonstrates the effectiveness of AARF. The average cumulative reward of the DQN with DVLS is beyond eight percent, and the convergence speed is improved by five percent compared to a classic DQN.

Mobile LiDAR-based Real-time Identification of Transmission Lines

Abstract. This paper proposes a method for identifying 3D point cloud of transmission line acquired by light detection and ranging (LiDAR) real-time mobile scanning. Since the single frame of point cloud obtained by LiDAR is sparse, the method employs a sliding spatial window strategy with Kalman filtering for dynamic point cloud registration. Then, a 3D point cloud deep learning neural network that utilizes uniform sampling and local feature aggregation (LFA) is designed specifically for transmission line objects. The network handles the problem of long-span objects and a large amount of point cloud. Finally, the instantiated transmission line objects are extracted from the top-down projection of the semantically segmented 3D point cloud by fast Euclidean clustering algorithm. Experiments demonstrate that the method achieves a classification accuracy of 94.7% and a mean intersection over union of 81.6% on 3D point cloud datasets of transmission line obtained from LiDAR mobile scanning, validating its ability to achieve real-time identification and distance measurement of transmission line objects.

Real-Time Detection Technology of Corn Kernel Breakage and Mildew Based on Improved YOLOv5s

In order to solve low recognition of corn kernel breakage degree and corn kernel mildew degree during corn kernel harvesting, this paper proposes a real-time detection method for corn kernel breakage and mildew based on improved YOlOv5s, which is referred to as the CST-YOLOv5s model algorithm in this paper. The method continuously obtains images through the discrete uniform sampling device of corn kernels and generates whole corn kernels, breakage corn kernels, and mildew corn kernel dataset samples. We aimed at the problems of high similarity of some corn kernel features in the acquired images and the low precision of corn kernel breakage and mildew recognition. Firstly, the CBAM attention mechanism is added to the backbone network of YOLOv5s to finely allocate and process the feature information, highlighting the features of corn breakage and mildew. Secondly, the pyramid pooling structure SPPCPSC, which integrates cross-stage local networks, is adopted to replace the SPPF in YOLOv5s. SPP and CPSC technologies are used to extract and fuse features of different scales, improving the precision of object detection. Finally, the original prediction head is converted into a transformer prediction head to explore the prediction potential with a multi-head attention mechanism. The experimental results show that the CST-YOLOv5s model has a significant improvement in the detection of corn kernel breakage and mildew. Compared with the original YOLOv5s model, the average precision (AP) of corn kernel breakage and mildew recognition increased by 5.2% and 7.1%, respectively, and the mean average precision (mAP) of all kinds of corn kernel recognition is 96.1%, and the frame rate is 36.7 FPS. Compared with YOLOv4-tiny, YOLOv6n, YOLOv7, YOLOv8s, and YOLOv9-E detection model algorithms, the CST-YOLOv5s model has better overall performance in terms of detection accuracy and speed. This study can provide a reference for real-time detection of breakage and mildew kernels during the harvesting process of corn kernels.