Count Models Research Articles

Segmentation and counting of wheat spike grains based on deep learning and textural feature

BackgroundGrain count is crucial to wheat yield composition and estimating yield parameters. However, traditional manual counting methods are time-consuming and labor-intensive. This study developed an advanced deep learning technique for the segmentation counting model of wheat grains. This model has been rigorously tested on three distinct wheat varieties: ‘Bainong 307’, ‘Xinmai 26’, and ‘Jimai 336’, and it has achieved unprecedented predictive counting accuracy.MethodThe images of wheat ears were taken with a smartphone at the late stage of wheat grain filling. We used image processing technology to preprocess and normalize the images to 480*480 pixels. A CBAM-HRNet wheat grain segmentation counting deep learning model based on the Convolutional Block Attention Module (CBAM) was constructed by combining deep learning, migration learning, and attention mechanism. Image processing algorithms and wheat grain texture features were used to build a grain counting and predictive counting model for wheat grains.ResultsThe CBAM-HRNet model using the CBAM was the best for wheat grain segmentation. Its segmentation accuracy of 92.04%, the mean Intersection over Union (mIoU) of 85.21%, the category mean pixel accuracy (mPA) of 91.16%, and the recall rate of 91.16% demonstrate superior robustness compared to other models such as HRNet, PSPNet, DeeplabV3+ , and U-Net. Method I for spike count, which calculates twice the number of grains on one side of the spike to determine the total number of grains, demonstrates a coefficient of determination R2 of 0.85, a mean absolute error (MAE) of 1.53, and a mean relative error (MRE) of 2.91. In contrast, Method II for spike count involves summing the number of grains on both sides to determine the total number of grains, demonstrating a coefficient of determination R2 of 0.92, an MAE) of 1.15, and an MRE) of 2.09%.ConclusionsImage segmentation algorithm of the CBAM-HRNet wheat spike grain is a powerful solution that uses the CBAM to segment wheat spike grains and obtain richer semantic information. This model can effectively address the challenges of small target image segmentation and under-fitting problems in training. Additionally, the spike grain counting model can quickly and accurately predict the grain count of wheat, providing algorithmic support for efficient and intelligent wheat yield estimation.

Semi-Supervised Crowd Counting With Spatial Temporal Consistency and Pseudo-Label Filter

Semi-supervised crowd counting (SSCC) aims to learn a crowd counting model with limited labeled images and a large number of unlabeled images. Previous works leverage unlabeled images by pseudo-labeling and spatial consistency regularization paradigms, which frequently adopt teacher-student frameworks. However, their performances are readily degraded due to the inconsistent and unreliable pseudo density map in complex crowd scenes. Here, we argue that the SSCC performance can be significantly improved by reducing the over-fitting of the incorrect pseudo labels, and a novel spatial-temporal consistency framework, named STC-Crowd, is proposed. Under different spatial perturbations, spatial consistency enables the counting model to output consistent predictions for the same crowd image. Temporal consistency generates similar feature embedding over adjacent training stages, alleviating the inconsistent issues of pseudo density maps generated for the same image over time. To store the temporal feature embedding of different density levels for temporal consistency, a dynamic temporal knowledge memory (DTKM) is deliberately designed, and considerably reduces the storage cost. Besides, a pseudo-label filter (PLF) mechanism is used to alleviate the negative impact of incorrect pseudo density maps, by reducing the supervision weights of unreliable pseudo labels with high uncertainty. Extensive experiments on four benchmark datasets show that our method obtains competent performance against leading SSCC methods, and especially works better on limited labeled images.

Improving performance of hurdle models using rare-event weighted logistic regression: an application to maternal mortality data.

In this paper, performance of hurdle models in rare events data is improved by modifying their binary component. The rare-event weighted logistic regression model is adopted in place of logistic regression to deal with class imbalance due to rare events. Poisson Hurdle Rare Event Weighted Logistic Regression (REWLR) and Negative Binomial Hurdle (NBH) REWLR are developed as two-part models which use the REWLR model to estimate the probability of a positive count and a Poisson or NB zero-truncated count model to estimate non-zero counts. This research aimed to develop and assess the performance of the Poisson and Negative Binomial (NB) Hurdle Rare Event Weighted Logistic Regression (REWLR) models, applied to simulated data with various degrees of zero inflation and to Nairobi county's maternal mortality data. The study data on maternal mortality were pulled from JPHES. The data contain the number of maternal deaths, which is the outcome variable, and other obstetric and demographic factors recorded in MNCH facilities in Nairobi between October 2021 and January 2022. The models were also fit and evaluated based on simulated data with varying degrees of zero inflation. The obtained results are numerically validated and then discussed from both the mathematical and the maternal mortality perspective. Numerical simulations are also presented to give a more complete representation of the model dynamics. Results obtained suggest that NB Hurdle REWLR is the best performing model for zero inflated count data due to rare events.

An in-depth analysis of head-on crash severity and fatalities in Ghana

Head-on collisions are often linked to more serious injuries compared to other types of crashes, due to the intense impact they cause. In low- and middle-income countries, these collisions frequently involve high occupancy public transportation vehicles, leading to higher fatality rates per crash. Given the high risk of injury and potential for multiple casualties, this study delves into the factors influencing the outcomes of head-on crashes and the number of fatalities in Ghana. The study analyzed six years of historical head-on collision data from Ghana and developed two models to address the issue. The injury-severity analysis was performed using a random parameter multinomial logit with heterogeneity in means and variances approach and aimed to identify the factors that have a significant impact on the severity of injuries sustained in head-on collisions, while the random parameters negative binomial fatality count model was designed to examine the factors that contribute to the number of fatalities in these crashes in the country. Results showed that head-on collisions with drivers over 65, buses, motorcycles, and those between 25 and 65 years of age were more likely to result in fatalities. Speeding and vehicle malfunctions were also found to be significant contributing factors to fatal head-on collisions. Head-on crashes involving minibuses and incidents where the driver was attempting to overtake another vehicle were found to be more likely to result in a higher number of fatalities. The results of this study uncover an intriguing interaction between human-related elements and socioeconomic factors, which pose obstacles to the Government's endeavor to upgrade the major highways in the country. Additionally, the increasing need for transportation has led to the presence of vehicles on the roads that may not meet safety standards. Consequently, it is no surprise that several of the study's findings align with expectations. Nevertheless, within the specific context of Ghana, these findings furnish compelling data-driven evidence supporting the adoption and implementation of the safe systems approach as a means to tackle fatal head-on collisions in the country.

Trap-Based Pest Counting: Multiscale and Deformable Attention CenterNet Integrating Internal LR and HR Joint Feature Learning

Pest counting, which predicts the number of pests in the early stage, is very important because it enables rapid pest control, reduces damage to crops, and improves productivity. In recent years, light traps have been increasingly used to lure and photograph pests for pest counting. However, pest images have a wide range of variability in pest appearance owing to severe occlusion, wide pose variation, and even scale variation. This makes pest counting more challenging. To address these issues, this study proposes a new pest counting model referred to as multiscale and deformable attention CenterNet (Mada-CenterNet) for internal low-resolution (LR) and high-resolution (HR) joint feature learning. Compared with the conventional CenterNet, the proposed Mada-CenterNet adopts a multiscale heatmap generation approach in a two-step fashion to predict LR and HR heatmaps adaptively learned to scale variations, that is, changes in the number of pests. In addition, to overcome the pose and occlusion problems, a new between-hourglass skip connection based on deformable and multiscale attention is designed to ensure internal LR and HR joint feature learning and incorporate geometric deformation, thereby resulting in improved pest counting accuracy. Through experiments, the proposed Mada-CenterNet is verified to generate the HR heatmap more accurately and improve pest counting accuracy owing to multiscale heatmap generation, joint internal feature learning, and deformable and multiscale attention. In addition, the proposed model is confirmed to be effective in overcoming severe occlusions and variations in pose and scale. The experimental results show that the proposed model outperforms state-of-the-art crowd counting and object detection models.

Family firm heterogeneity and patenting. Revising the role of size and age

This study offers a comprehensive framework for evaluating the moderating effects of size and age on the relationship between family ownership and innovation. The research hypotheses are tested on a large sample of Italian firms observed over the 2010–2017 period using a zero-inflated nonlinear count model. To refine the understanding of firm heterogeneity, we employ a three-way interaction approach. Results show that the patenting gap between FFs and non-FFs is sensitive to size and age. Compared to non-FFs, FFs underperform when they are small and young, or large and mature, with no substantial differences seen in other types of firms. By referring to the Socioemotional Wealth theory, we find that the founder effect, which differs over a firm’s life, is behind a good deal of these findings.

A Review of Single-Cell RNA-Seq Annotation, Integration, and Cell-Cell Communication.

Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful tool for investigating cellular biology at an unprecedented resolution, enabling the characterization of cellular heterogeneity, identification of rare but significant cell types, and exploration of cell-cell communications and interactions. Its broad applications span both basic and clinical research domains. In this comprehensive review, we survey the current landscape of scRNA-seq analysis methods and tools, focusing on count modeling, cell-type annotation, data integration, including spatial transcriptomics, and the inference of cell-cell communication. We review the challenges encountered in scRNA-seq analysis, including issues of sparsity or low expression, reliability of cell annotation, and assumptions in data integration, and discuss the potential impact of suboptimal clustering and differential expression analysis tools on downstream analyses, particularly in identifying cell subpopulations. Finally, we discuss recent advancements and future directions for enhancing scRNA-seq analysis. Specifically, we highlight the development of novel tools for annotating single-cell data, integrating and interpreting multimodal datasets covering transcriptomics, epigenomics, and proteomics, and inferring cellular communication networks. By elucidating the latest progress and innovation, we provide a comprehensive overview of the rapidly advancing field of scRNA-seq analysis.

Bayesian Inference for a New Negative Binomial-Samade Model for Time Series Data Counts with Its Properties and Applications

A new distribution was developed that mixed the negative binomial (NB) and Samade distributions, called the negative binomial-Samade (NB-SA) distribution. The properties of this distribution were studied, and the newly created distribution was applied using the framework of generalized linear models to build a time series data count model. The characteristics of overdispersion and heavy-tailed distribution of the count response variables were applied in the actual dataset modeling. Distribution parameters and the regression coefficient were estimated using a Bayesian approach. Results showed that the NB-SA model had significantly the highest efficiency compared with the classical NB and Poisson models for analyzing factors influencing the daily number of COVID-19 deaths in Thailand.

Tomato Maturity Detection and Counting Model Based on MHSA-YOLOv8.

The online automated maturity grading and counting of tomato fruits has a certain promoting effect on digital supervision of fruit growth status and unmanned precision operations during the planting process. The traditional grading and counting of tomato fruit maturity is mostly done manually, which is time-consuming and laborious work, and its precision depends on the accuracy of human eye observation. The combination of artificial intelligence and machine vision has to some extent solved this problem. In this work, firstly, a digital camera is used to obtain tomato fruit image datasets, taking into account factors such as occlusion and external light interference. Secondly, based on the tomato maturity grading task requirements, the MHSA attention mechanism is adopted to improve YOLOv8's backbone to enhance the network's ability to extract diverse features. The Precision, Recall, F1-score, and mAP50 of the tomato fruit maturity grading model constructed based on MHSA-YOLOv8 were 0.806, 0.807, 0.806, and 0.864, respectively, which improved the performance of the model with a slight increase in model size. Finally, thanks to the excellent performance of MHSA-YOLOv8, the Precision, Recall, F1-score, and mAP50 of the constructed counting models were 0.990, 0.960, 0.975, and 0.916, respectively. The tomato maturity grading and counting model constructed in this study is not only suitable for online detection but also for offline detection, which greatly helps to improve the harvesting and grading efficiency of tomato growers. The main innovations of this study are summarized as follows: (1) a tomato maturity grading and counting dataset collected from actual production scenarios was constructed; (2) considering the complexity of the environment, this study proposes a new object detection method, MHSA-YOLOv8, and constructs tomato maturity grading models and counting models, respectively; (3) the models constructed in this study are not only suitable for online grading and counting but also for offline grading and counting.

VBASS enables integration of single cell gene expression data in Bayesian association analysis of rare variants

Rare or de novo variants have substantial contribution to human diseases, but the statistical power to identify risk genes by rare variants is generally low due to rarity of genotype data. Previous studies have shown that risk genes usually have high expression in relevant cell types, although for many conditions the identity of these cell types are largely unknown. Recent efforts in single cell atlas in human and model organisms produced large amount of gene expression data. Here we present VBASS, a Bayesian method that integrates single-cell expression and de novo variant (DNV) data to improve power of disease risk gene discovery. VBASS models disease risk prior as a function of expression profiles, approximated by deep neural networks. It learns the weights of neural networks and parameters of Gamma-Poisson likelihood models of DNV counts jointly from expression and genetics data. On simulated data, VBASS shows proper error rate control and better power than state-of-the-art methods. We applied VBASS to published datasets and identified more candidate risk genes with supports from literature or data from independent cohorts. VBASS can be generalized to integrate other types of functional genomics data in statistical genetics analysis.

Evaluation of Bayesian spatiotemporal infectious disease models for prospective surveillance analysis

BackgroundCOVID-19 brought enormous challenges to public health surveillance and underscored the importance of developing and maintaining robust systems for accurate surveillance. As public health data collection efforts expand, there is a critical need for infectious disease modeling researchers to continue to develop prospective surveillance metrics and statistical models to accommodate the modeling of large disease counts and variability. This paper evaluated different likelihoods for the disease count model and various spatiotemporal mean models for prospective surveillance.MethodsWe evaluated Bayesian spatiotemporal models, which are the foundation for model-based infectious disease surveillance metrics. Bayesian spatiotemporal mean models based on the Poisson and the negative binomial likelihoods were evaluated with the different lengths of past data usage. We compared their goodness of fit and short-term prediction performance with both simulated epidemic data and real data from the COVID-19 pandemic.ResultsThe simulation results show that the negative binomial likelihood-based models show better goodness of fit results than Poisson likelihood-based models as deemed by smaller deviance information criteria (DIC) values. However, Poisson models yield smaller mean square error (MSE) and mean absolute one-step prediction error (MAOSPE) results when we use a shorter length of the past data such as 7 and 3 time periods. Real COVID-19 data analysis of New Jersey and South Carolina shows similar results for the goodness of fit and short-term prediction results. Negative binomial-based mean models showed better performance when we used the past data of 52 time periods. Poisson-based mean models showed comparable goodness of fit performance and smaller MSE and MAOSPE results when we used the past data of 7 and 3 time periods.ConclusionWe evaluate these models and provide future infectious disease outbreak modeling guidelines for Bayesian spatiotemporal analysis. Our choice of the likelihood and spatiotemporal mean models was influenced by both historical data length and variability. With a longer length of past data usage and more over-dispersed data, the negative binomial likelihood shows a better model fit than the Poisson likelihood. However, as we use a shorter length of the past data for our surveillance analysis, the difference between the Poisson and the negative binomial models becomes smaller. In this case, the Poisson likelihood shows robust posterior mean estimate and short-term prediction results.

An Extension of the Geometric Distribution with Properties and Applications

A new parameter is introduced to extend the geometric distribution using Azzalini's method. Several important structural properties of the proposed two-parameter extended geometric distribution are investigated. Characterizations including for the geometric distribution, in terms of the proposed model, are established. Maximum likelihood estimation, method of moment estimation and relative frequency based estimation of the parameters are discussed in detail. The likelihood ratio test regarding relevance of the additional parameter is presented. Bayesian estimation of the parameters using STAN is also discussed. The proposed model is compared with some recently introduced two-parameter count models by analyzing two real-life datasets. The findings clearly indicate superiority of the proposed model over the rest.

Clinical study on the evaluation of liver fibrosis by ultrasound elastography combined with platelet count model.

To establish the prediction model of liver fibrosis by combining ultrasound elastography and platelet count and evaluates its clinical value. 146 patients with chronic liver diseases(CLD) admitted to our hospital from July 2020 to July 2022 were collected for liver biopsy pathological examination, and the results of ultrasound elastography (liver hardness value) and serological indicators were collected. Based on the results of Spearman correlation test and multiple linear regression model, the prediction model of liver fibrosis using ultrasound elastography combined with platelet count was constructed and verified. The AUC of transient elastography combined with platelet count model (FSP) in the diagnosis of S2, S3 and S4 phases of liver fibrosis was 0.665, 0.835 and 0.909, with specificity of 81.5%, 90.0% and 100%. The AUC of sound touch elastography combined with platelet count model (STEP) in diagnosing S2, S3 and S4 phases of liver fibrosis was 0.685, 0.810 and 0884, with specificity of 96.3%, 90.0% and 83.3%, which are higher than APRI, FIB-4, FORNS, AAR and other models. Ultrasound elastography combined with platelet count model has good diagnostic efficacy for liver fibrosis.

Strawberry R-CNN: Recognition and counting model of strawberry based on improved faster R-CNN

Computer technology's rapid advancement has significantly enhanced global agricultural modernization, notably improving agricultural production efficiency. Given the vulnerability of strawberries during the harvesting process, the automatic harvesting technology for strawberries necessitates a highly accurate recognition algorithm. In this paper, we introduce a model, Strawberry R-CNN, designed for intelligent recognition and counting of strawberries in natural environments. The Strawberry R-CNN model enhances strawberry recognition accuracy by refining the Faster R-CNN through several key modifications. Firstly, we replaced the VGG16 in the original Faster R-CNN with an improved multi-cascade network structure for feature extraction. This change allows for the capture of rich location data and detailed information typically absent in higher-level features. Secondly, the RoiPooling operation was replaced with RoiAlign to eliminate the error associated with the rounding method in two quantization stages. Lastly, we utilized a bilinear interpolation method for computation, preserving the floating-point number, and reducing model error. For strawberry counting, we proposed an efficient and practical evaluation method by creating an error set for strawberry counting. Experimental results demonstrated that the Strawberry R-CNN model achieved an average precision (AP) of 0.9019 for ripe strawberries and 0.8447 for immature ones, with a mean average precision (mAP) of 0.8733. The counting accuracy for ripe and immature strawberries was 99.1% and 73.7% respectively. The method presented in this work exhibits strong detection and counting capabilities, suitable for automatic monitoring, harvesting, and yield estimation of strawberries.

Crash Frequency Minimization with Severity Mitigation in Road Geometric Design Using Chance Constraint Programming Optimization

Crash frequency and crash severity are two major aspects of transportation safety. In this paper, we propose a decision-making scheme combining statistical analysis and optimization modeling to be used in transportation safety study. We conduct a safety analysis, a travel speed analysis, and an optimization analysis to develop a two-stage decision scheme to minimize crash frequency while mitigating crash severity, using data collected in urban environments in Lincoln, Nebraska. In the safety analysis and the travel speed analysis, we study the impact of lane width and other related road geometric design parameters on annual crash frequency and vehicle travel speed using count models and linear regression models, respectively. In the optimization analysis, the proposed two-stage stochastic programming model determines the lane width and other road geometric design parameters in the first stage, and then the posted road speed limit in the second stage for each scenario. To mitigate crash severity, we use a chance constraint to restrict a certain percentile of the vehicle travel speed to comply with the posted road speed limit. This two-stage decision scheme is shown to be effective for the data collected in Lincoln, Nebraska, when restricting the vehicle travel speed of up to two positive standard deviations from the mean travel speed to be under the posted speed limit. The application of a stochastic programming model that utilizes regression analysis results serves as an innovative decision scheme that effectively connects statistical analysis and optimization studies in road geometric design for transportation safety. Its objective is to minimize crash frequency while simultaneously mitigating crash severity. This methodology has extensive potential for application in various environments to assist in the reduction of both crash frequency and crash severity.

Investigating the temporal differences among bike-sharing users through comparative analysis based on count, time series, and data mining models

Bike-sharing services provide easy access to environmentally-friendly mobility reducing congestion in urban areas. Increasing demand requires highly service planning methods based on bike-sharing user behavior. Negative Binomial, Poisson Regression, and Time Series models were elaborated considering the weather to reveal the differences between the members, occasional users, and visitor bike-sharing user groups. The negative Binomial approach is found to be superior to Poisson. Weather effects were varied in their influence on bike-sharing user classifications. In general, good weather conditions lead to more usage of bike-sharing. Weekends attract more occasional users and visitors than weekdays. In time series models, the seasonal trend of bike-sharing trips conducted by members was predicted without weather impact. According to the comparison, Random Forest performed better than SARIMA when the number of observations was low. Visitors are more influenced by temperature, wind and type of day. Occasional users are more subjected to precipitation. For members, it is found that the temperature, type of day are the most significant factors. The least factors for all are varied as well: precipitation for visitors, humidity for occasional users, precipitation and wind for members. The results help decision-makers predict the daily usage of bike-sharing for various user groups.

Mortality Caused by Candida auris Bloodstream Infections in Comparison with Other Candida Species, a Multicentre Retrospective Cohort.

Candida auris is an emerging pathogen considered to be critical in the World Health Organization fungal organisms list. The study aims to determine the mortality and hospital stays attributed to Candida auris (C. auris) compared to other Candida species in adult patients with candidemia. A retrospective cohort of adults with candidemia was examined from seven centres in Colombia between 2016 and 2021. The primary outcome was 30-day mortality, and the secondary outcome was the length of hospital stay among survivors. Adjustment of the confounding variables was performed using inverse probability weights of exposure propensity score (candidemia by C. auris), survival regression models (Weibull distribution), and a counting model (negative binomial distribution). A value of 244 (47.6%) of the 512 patients with candidemia died within the first 30 days. The crude mortality in C. auris was 38.1% vs. 51.1% in Candida non-auris (CNA). In the Weibull model, mortality in the C. auris group was lower (adjusted HR: aHR- 0.69, 95% CI: 0.53-0.90). Antifungal treatment also decreased mortality, with an aHR of 0.36 (95% CI 0.27-0.47), while the presence of septic shock on patient progression increased it, with an aHR of 1.73 (95% CI 1.41-2.13). Among the patients who survived, no differences in the length of hospital stay were observed between the C. auris and the CNA groups, with an incidence rate ratio of 0.92 (95% CI: 0.68-1.22). Mortality in patients with C. auris bloodstream infections appears lower when adjusted for numerous confounding variables regarding treatment and the presence of septic shock in patient progression. We identified no significant effect of C. auris on the length of hospital stay in surviving patients.

Recombination and sterility in inversion homo- and heterokaryotypes under a general counting model of chiasma interference.

It has long been known that the chiasmata are not independently distributed in most organisms, a phenomenon known as chiasma interference. In this paper, I suggest a model of chiasma interference that generalizes the Poisson model, the counting model, the Poisson-skip model and the two-pathway counting model into a single framework, and use it to derive infinite series expressions for the sterility and recombination pattern probabilities in inversion homo- and heterokaryotypes, and a closed-form expression for the special case of the two-pathway counting model in homokaryotypes. I then use these expressions to perform maximum likelihood parameter estimations for recombination and tetrad data from various species. The results imply that the simpler counting models perform well compared to more complex ones, that interference works in a similar way in homo- and heterokaryotypes, and that the model fits well with data for the latter as well as the former. I also find evidence that the interference signal is broken by the centromere in some species, but not others, suggestions of negative interference in Aspergillus nidulans, and no consistent support for the theory that a second non-interfering chiasma pathway exists only in organisms that require double-strand break for synapsis. I suggest that the latter finding is at least partly due to issues involved in analysing aggregate data from different experiments and individuals.

Automated Vehicle Counting from Pre-Recorded Video Using You Only Look Once (YOLO) Object Detection Model.

Different techniques are being applied for automated vehicle counting from video footage, which is a significant subject of interest to many researchers. In this context, the You Only Look Once (YOLO) object detection model, which has been developed recently, has emerged as a promising tool. In terms of accuracy and flexible interval counting, the adequacy of existing research on employing the model for vehicle counting from video footage is unlikely sufficient. The present study endeavors to develop computer algorithms for automated traffic counting from pre-recorded videos using the YOLO model with flexible interval counting. The study involves the development of algorithms aimed at detecting, tracking, and counting vehicles from pre-recorded videos. The YOLO model was applied in TensorFlow API with the assistance of OpenCV. The developed algorithms implement the YOLO model for counting vehicles in two-way directions in an efficient way. The accuracy of the automated counting was evaluated compared to the manual counts, and was found to be about 90 percent. The accuracy comparison also shows that the error of automated counting consistently occurs due to undercounting from unsuitable videos. In addition, a benefit-cost (B/C) analysis shows that implementing the automated counting method returns 1.76 times the investment.

Domain-General Crowd Counting in Unseen Scenarios

Domain shift across crowd data severely hinders crowd counting models to generalize to unseen scenarios. Although domain adaptive crowd counting approaches close this gap to a certain extent, they are still dependent on the target domain data to adapt (e.g. finetune) their models to the specific domain. In this paper, we instead target to train a model based on a single source domain which can generalize well on any unseen domain. This falls into the realm of domain generalization that remains unexplored in crowd counting. We first introduce a dynamic sub-domain division scheme which divides the source domain into multiple sub-domains such that we can initiate a meta-learning framework for domain generalization. The sub-domain division is dynamically refined during the meta-learning. Next, in order to disentangle domain-invariant information from domain-specific information in image features, we design the domain-invariant and -specific crowd memory modules to re-encode image features. Two types of losses, i.e. feature reconstruction and orthogonal losses, are devised to enable this disentanglement. Extensive experiments on several standard crowd counting benchmarks i.e. SHA, SHB, QNRF, and NWPU, show the strong generalizability of our method. Our code is available at: https://github.com/ZPDu/Domain-general-Crowd-Counting-in-Unseen-Scenarios