Counting Datasets Research Articles

Analyzing Number of Children Ever Born in Bangladesh Using Generalized Poisson Regression Model

One major indicator to detect any country’s future population size is the total fertility rate (TFR). Based on the total number of children ever born, an underdispersed count data, this paper aims to investigate the fertility behavior of mothers in Bangladesh who have given birth to one or more children. For analyzing count dataset, the Poisson regression model (PRM) is commonly used, which assumes that the response variable’s mean is equal to its variance. Since the generalized Poisson regression model (GPRM) is suitable for the analysis of both overdispersed and underdispersed count data, this model is applied to deal with the underlying count dataset in this study. The results obtained in this paper have revealed some variables that impose significant impact on the current fertility differential of mothers giving birth to children. Dhaka Univ. J. Sci. 72(2): 01-06, 2024 (July)

Object Counting via Group and Graph Attention Network.

Object counting, defined as the task of accurately predicting the number of objects in static images or videos, has recently attracted considerable interest. However, the unavoidable presence of background noise prevents counting performance from advancing further. To address this issue, we created a group and graph attention network (GGANet) for dense object counting. GGANet is an encoder-decoder architecture incorporating a group channel attention (GCA) module and a learnable graph attention (LGA) module. The GCA module groups the feature map into several subfeatures, each of which is assigned an attention factor through the identical channel attention. The LGA module views the feature map as a graph structure in which the different channels represent diverse feature vertices, and the responses between channels represent edges. The GCA and LGA modules jointly avoid the interference of irrelevant pixels and suppress the background noise. Experiments are conducted on four crowd-counting datasets, two vehicle-counting datasets, one remote-sensing counting dataset, and one few-shot object-counting dataset. Comparative results prove that the proposed GGANet achieves superior counting performance.

Classifying Anemia Diseases based on VGG-16 and CBAM Attention Mechanism Models

Anemia is a major public health concern and has a significant impact on women and children worldwide. This condition is caused by a lack of sufficient Red Blood Cells (RBCs). Traditionally, anemia diagnosis has relied on invasive techniques that require blood samples, leading to pain and discomfort. This study explores deep learning techniques for automating and enhancing anemia detection accuracy using the Blood Cell Count and Detection (BCCD) dataset. We employ the VGG16 convolutional neural network architecture augmented with the Convolutional Block Attention Module (CBAM) to boost feature representation and performance. The BCCD dataset, consisting of 364 images with 4888 labeled blood cells, was used for training and evaluation. Our enhanced VGG16 model achieved accuracy for RBCs, WBCs, and platelets with values of 0.84, 0.93, and 0.92, respectively, effectively identifying various blood cell types and detecting anemia. Precision-recall analysis and confusion matrix metrics confirmed the robustness of the model, with high precision and recall rates and minimal false positives and negatives. These results suggest that advanced deep learning models with attention mechanisms, such as CBAM, can significantly improve medical diagnostics by providing reliable, efficient, and accurate early disease detection tools. Future development will be directed towards fine-tuning and validating the proposed model on other medical imaging datasets for clinical applications.

Vision foundation model for agricultural applications with efficient layer aggregation network

Agricultural production is transitioning from traditional tools to IoT-connected automation devices. The integration of computer vision and agricultural automation is becoming closer with the rapid development of deep learning technology. Agricultural vision tasks, nevertheless, are relatively simple and less diverse compared to large-scale vision tasks, and there is a scarcity of plant data. This is also the reason why researchers need to make significant adjustments when directly applying advanced algorithms to the agricultural domain. In practice, downstream vision models should be designed to be concise and efficient. In this paper, we meticulously designed a deep convolutional network called TasselELANet for agricultural vision applications. It features a concise and efficient global architecture, including a 16-fold downsampling layer encoder and a decoder utilizing only 2 feature layers, which is greatly different from the general approaches. At its core is the efficient layer aggregation network, which utilizes the cross-stage fusion strategy to effectively optimize the gradient propagation path, thereby enhancing the learning ability and inference speed of the network. We validated the performance of TasselELANet on three challenging publicly available plant detection and counting datasets: maize tassels, wheat ears, and rice panels. The experimental tests achieved an average Accuracy of 0.899 and an average determination coefficient R2 of 0.867, remarkably outperforming other advanced computer vision methods in terms of accuracy and efficiency, and demonstrating sufficient generalizability. We firmly believe that TasselELANet can serve as a powerful and reliable vision tool for agricultural practitioners while also providing a solid foundation for future research. The code, datasets can be accessed at https://github.com/Ye-Sk/TasselELANet.git.

CrowdUNet: Segmentation assisted U-shaped crowd counting network

With the end of the COVID-19 pandemic, the number of pedestrians in various public places has increased dramatically. Estimating the size and density distribution of crowds accurately from images is essential for public safety. At present, there are still many factors that limit the accuracy of dense crowd counting, such as perspective distortion, background clutter and heavy occlusion. To be capable of accurately estimating the crowd size in RGB images, we propose a crowd counting network called CrowdUNet, which is assisted by a segmentation task. It applies the segmentation results to the crowd counts, making the network more focused on the prediction of foreground regions. We combine Swin Transformer Block and CNN to build a Swin Transformer Convolution(STC) module to extract deep semantic features. We analyze the characteristics of crowd images and propose a novel decoder structure called Coordinate Decoder(CD), which better aggregates low and high level features and improve the robustness of the network. In order to obtain accurate regression results, we also propose a regression head with multi-scale receptive fields, which is called Spatial Pyramid Convolution (SPC). Extensive experiments on four challenging crowd counting datasets namely ShanghaiTech A, ShanghaiTech B, UCF p=CC 50, and UCF-QNRF have validated the proposed method.

AdaTreeFormer: Few shot domain adaptation for tree counting from a single high-resolution image

The process of estimating and counting tree density using only a single aerial or satellite image is a difficult task in the fields of photogrammetry and remote sensing. However, it plays a crucial role in the management of forests. The huge variety of trees in varied topography severely hinders tree counting models to perform well. The purpose of this paper is to propose a framework that is learnt from the source domain with sufficient labeled trees and is adapted to the target domain with only a limited number of labeled trees. Our method, termed as AdaTreeFormer, contains one shared encoder with a hierarchical feature extraction scheme to extract robust features from the source and target domains. It also consists of three subnets: two for extracting self-domain attention maps from source and target domains respectively and one for extracting cross-domain attention maps. For the latter, an attention-to-adapt mechanism is introduced to distill relevant information from different domains while generating tree density maps; a hierarchical cross-domain feature alignment scheme is proposed that progressively aligns the features from the source and target domains. We also adopt adversarial learning into the framework to further reduce the gap between source and target domains. Our AdaTreeFormer is evaluated on six designed domain adaptation tasks using three tree counting datasets, i.e. Jiangsu, Yosemite, and London. Experimental results show that AdaTreeFormer significantly surpasses the state of the art, e.g. in the cross domain from the Yosemite to Jiangsu dataset, it achieves a reduction of 15.9 points in terms of the absolute counting errors and an increase of 10.8% in the accuracy of the detected trees’ locations. The codes and datasets are available at https://github.com/HAAClassic/AdaTreeFormer.

CC-DETR: DETR with Hybrid Context and Multi-Scale Coordinate Convolution for Crowd Counting

Prevailing crowd counting approaches primarily rely on density map regression methods. Despite wonderful progress, significant scale variations and complex background interference within the same image remain challenges. To address these issues, in this paper we propose a novel DETR-based crowd counting framework called Crowd Counting DETR (CC-DETR), which aims to extend the state-of-the-art DETR object detection framework to the crowd counting task. In CC-DETR, a DETR-like encoder–decoder structure (Hybrid Context DETR, i.e., HCDETR) is proposed to tackle complex visual information by fusing features from hybrid semantic levels through a transformer. In addition, we design a Coordinate Dilated Convolution Module (CDCM) to effectively employ position-sensitive context information in different scales. Extensive experiments on three challenging crowd counting datasets (ShanghaiTech, UCF-QNRF, and NWPU) demonstrate that our model is effective and competitive when compared against SOTA crowd counting models.

Late better than early: A decision-level information fusion approach for RGB-Thermal crowd counting with illumination awareness

In this paper, we make the first research effort to address the RGB-Thermal (RGB-T) crowd counting problem with the decision-level late fusion manner. Being different from the existing pixel-level or feature-level fusion methods, our proposition chooses to fuse the density maps yielded by RGB and thermal counterparts via spatially adaptive weighting with RGB illumination-aware attention. Our key intuition to conduct RGB-T density map fusion lies in 2 main folders. First, compared with the raw RGB-T images or convolutional feature maps, RGB-T density maps contain stronger counting-wise semantic meanings. Secondly, they are also of high spatial resolution for revealing fine local details. To fuse them adaptively, a spatial weighting map for each modality, together with an illumination-related RGB weight is generated. In this way, the issues of RGB illumination awareness and local counting pattern characterization ability are concerned jointly. To the best of our knowledge, we are the first to leverage RGB-T crowd counting concerning these 2 issues in a unified way. Meanwhile, cross-modality feature interaction is conducted between RGB and thermal modalities to facilitate spatial weighting map generation. The experiments on 2 well-established RGB-T crowd counting datasets (i.e., RGBT-CC and DroneRGBT) verify the superiority of our proposition. The source code and pretrained models will be released upon acceptance at https://github.com/hustaia/DLF-IA.

A simulation study to assess the impact missing values on the performance of different statistical methods for analysis of binary repeated measures data with an additional hierarchical structure

The primary objective of the study was to assess the impact of missing values on the analy- sis of binary repeated measures data with an additional hierarchical structure. One motivat- ing example for the present study was records of high somatic cell counts in milk samples obtained by approximately monthly sampling throughout the lactations of cows in dairy herds. Random effects models with autocorrelated (ρ = 1, 0.9 or 0.5) subject-level ran- dom effects were behind the simulated data. In general, the settings of the simulation were chosen to reflect a real somatic cell count dataset (scc40), except that the within-cow time series length was set to 8-time points for each cow. The estimation procedures consid- ered were: Ordinary Logistic Regression (OLR), Alternating Logistic Regression (ALR), Weighted Generalized Estimating Equations (WGEE), Penalized Quasi Likelihood (PQL), Maximum likelihood via numerical integration (ML) and Bayesian Markov chain Monte Carlo (MCMC). Multiple scenarios of simulated incomplete datasets were considered and include: a scenario corresponded to a combination of missingness patterns present in the scc40 dataset (scc40 scenario) The remaining scenarios involved only drop-outs, and corre- sponded to either moderate or high percentages of values either missing at random (MAR) or not missing at random (NMAR), respectively. In the scc40 scenario, all estimation procedures except OLR performed well and produced estimates with small relative bias (generally less than 5%) for levels of missingness that roughly corresponded to the scc40 data. In MAR missingness scenarios, some biases were found for ALR, WGEE and PQL procedures, whereas the likelihood-based procedures were largely unaffected by the miss- ing values. In NMAR scenarios, all procedures experienced similar and strong biases in the time coefficient; however, fixed effects estimates at the subject and cluster levels were relatively unaffected. Journal of Statistical Research 2023, Vol 57, No.1-2, pp.35-67

Rice Counting and Localization in Unmanned Aerial Vehicle Imagery Using Enhanced Feature Fusion

In rice cultivation and breeding, obtaining accurate information on the quantity and spatial distribution of rice plants is crucial. However, traditional field sampling methods can only provide rough estimates of the plant count and fail to capture precise plant locations. To address these problems, this paper proposes P2PNet-EFF for the counting and localization of rice plants. Firstly, through the introduction of the enhanced feature fusion (EFF), the model improves its ability to integrate deep semantic information while preserving shallow spatial details. This allows the model to holistically analyze the morphology of plants rather than focusing solely on their central points, substantially reducing errors caused by leaf overlap. Secondly, by integrating efficient multi-scale attention (EMA) into the backbone, the model enhances its feature extraction capabilities and suppresses interference from similar backgrounds. Finally, to evaluate the effectiveness of the P2PNet-EFF method, we introduce the URCAL dataset for rice counting and localization, gathered using UAV. This dataset consists of 365 high-resolution images and 173,352 point annotations. Experimental results on the URCAL demonstrate that the proposed method achieves a 34.87% reduction in MAE and a 28.19% reduction in RMSE compared to the original P2PNet while increasing R2 by 3.03%. Furthermore, we conducted extensive experiments on three frequently used plant counting datasets. The results demonstrate the excellent performance of the proposed method.

Research on Dynamic Pig Counting Method Based on Improved YOLOv7 Combined with DeepSORT.

A pig inventory is a crucial component of achieving precise and large-scale farming. In complex pigsty environments, due to pigs' stress reactions and frequent obstructions, it is challenging to count them accurately and automatically. This difficulty contrasts with most current deep learning studies, which rely on overhead views or static images for counting. This research proposes a video-based dynamic counting method, combining YOLOv7 with DeepSORT. By utilizing the YOLOv7 network structure and optimizing the second and third 3 × 3 convolution operations in the head network ELAN-W with PConv, the model reduces the computational demand and improves the inference speed without sacrificing accuracy. To ensure that the network acquires accurate position perception information at oblique angles and extracts rich semantic information, we introduce the coordinate attention (CA) mechanism before the three re-referentialization paths (REPConv) in the head network, enhancing robustness in complex scenarios. Experimental results show that, compared to the original model, the improved model increases the mAP by 3.24, 0.05, and 1.00 percentage points for oblique, overhead, and all pig counting datasets, respectively, while reducing the computational cost by 3.6 GFLOPS. The enhanced YOLOv7 outperforms YOLOv5, YOLOv4, YOLOv3, Faster RCNN, and SSD in target detection with mAP improvements of 2.07, 5.20, 2.16, 7.05, and 19.73 percentage points, respectively. In dynamic counting experiments, the improved YOLOv7 combined with DeepSORT was tested on videos with total pig counts of 144, 201, 285, and 295, yielding errors of -3, -3, -4, and -26, respectively, with an average accuracy of 96.58% and an FPS of 22. This demonstrates the model's capability of performing the real-time counting of pigs in various scenes, providing valuable data and references for automated pig counting research.

Efficient Nonparametric Tensor Decomposition for Binary and Count Data

In numerous applications, binary reactions or event counts are observed and stored within high-order tensors. Tensor decompositions (TDs) serve as a powerful tool to handle such high-dimensional and sparse data. However, many traditional TDs are explicitly or implicitly designed based on the Gaussian distribution, which is unsuitable for discrete data. Moreover, most TDs rely on predefined multi-linear structures, such as CP and Tucker formats. Therefore, they may not be effective enough to handle complex real-world datasets. To address these issues, we propose ENTED, an Efficient Nonparametric TEnsor Decomposition for binary and count tensors. Specifically, we first employ a nonparametric Gaussian process (GP) to replace traditional multi-linear structures. Next, we utilize the Pólya-Gamma augmentation which provides a unified framework to establish conjugate models for binary and count distributions. Finally, to address the computational issue of GPs, we enhance the model by incorporating sparse orthogonal variational inference of inducing points, which offers a more effective covariance approximation within GPs and stochastic natural gradient updates for nonparametric models. We evaluate our model on several real-world tensor completion tasks, considering binary and count datasets. The results manifest both better performance and computational advantages of the proposed model.

A Fixed-Point Approach to Unified Prompt-Based Counting

Existing class-agnostic counting models typically rely on a single type of prompt, e.g., box annotations. This paper aims to establish a comprehensive prompt-based counting framework capable of generating density maps for concerned objects indicated by various prompt types, such as box, point, and text. To achieve this goal, we begin by converting prompts from different modalities into prompt masks without requiring training. These masks are then integrated into a class-agnostic counting methodology for predicting density maps. Furthermore, we introduce a fixed-point inference along with an associated loss function to improve counting accuracy, all without introducing new parameters. The effectiveness of this method is substantiated both theoretically and experimentally. Additionally, a contrastive training scheme is implemented to mitigate dataset bias inherent in current class-agnostic counting datasets, a strategy whose effectiveness is confirmed by our ablation study. Our model excels in prominent class-agnostic datasets and exhibits superior performance in cross-dataset adaptation tasks.

Gramformer: Learning Crowd Counting via Graph-Modulated Transformer

Transformer has been popular in recent crowd counting work since it breaks the limited receptive field of traditional CNNs. However, since crowd images always contain a large number of similar patches, the self-attention mechanism in Transformer tends to find a homogenized solution where the attention maps of almost all patches are identical. In this paper, we address this problem by proposing Gramformer: a graph-modulated transformer to enhance the network by adjusting the attention and input node features respectively on the basis of two different types of graphs. Firstly, an attention graph is proposed to diverse attention maps to attend to complementary information. The graph is building upon the dissimilarities between patches, modulating the attention in an anti-similarity fashion. Secondly, a feature-based centrality encoding is proposed to discover the centrality positions or importance of nodes. We encode them with a proposed centrality indices scheme to modulate the node features and similarity relationships. Extensive experiments on four challenging crowd counting datasets have validated the competitiveness of the proposed method. Code is available at https://github.com/LoraLinH/Gramformer.

Abstract 2311: Precise assessment of cancer cell growth and survival by artificial intelligence

Abstract Introduction: Cancer research requires accurate methods for measuring analytical parameters like cell culture confluence, cell count, colony numbers, viability, and motility. These methods must be unbiased and user-independent for reproducible data. Cell analytics involves manual processes (e.g., manual cell counting) or reagent-based approaches (e.g., viability kits). In recent years, semi-automated systems have been introduced that can either count cells, measure cell growth by density tracking, and/or determine cell viability. However, these methods are often time-consuming, require reagents and labeling, and may involve costly instrumentation. Artificial Intelligence (AI) has made strides in clinical and laboratory research, holding promise for swift integration into cancer research. Here, we present the development and validation of SnapCyte™, an AI performing accurate, unbiased, label- and reagent-free cell analytics from basic cell culture images, independent of specialized instrumentation. Materials and Methods: Cell images were generated using diverse cell lines (MCF7, PC3, Hela, Raji⋯) cultured in various vessels with and without treatments (Taxane, Cisplatin, Heatshock⋯). For the cell count dataset, adherent cells were detached using trypsin, stained with trypan blue and loaded into a standard hemocytometer. Images were acquired using the SnapCyte™ adaptor and diverse microscopes (Leica, Hund, Zeiss, Nikon) and cell phones. Multiple datasets of 500 annotated images each were created, with images masked for confluency or for singular live and dead cells by experienced scientists. Object segmentation utilized the UNet architecture for localizing cells in cell culture images and iterative training was applied to achieve the required accuracy. Results: After multiple training iterations, SnapCyte™ AI detection models achieved 99% precision for confluency and >95% precision and recall for cell count. SnapCyte™ surpassed standard methods (Crystal Violet, WST1, MTT, Presto blue, CyQuant, Incucyte®, and Bio-Rad TC20 cell counter), displaying high accuracy and smaller standard error variation than reagent-based assays. Compared to IncuCyte® Bio-Rad TC20, SnapCyte™ demonstrated similar accuracy and greater user-independent results. Furthermore, SnapCyte™ acquired data in under 10 minutes, with non-invasive measurements, allowing direct use of cells in downstream assays. Conclusion: We have developed and validated an AI model for advanced cell analytics. Our data show that the SnapCyte ™ AI is at par or better than existing reagent and instrument-based solutions in assessing cell confluency, number, and viability. This technology offers a fast, accurate, and unbiased cell analytics platform that is resistant to user variations, and independent of reagents and costly equipment. Citation Format: Cheung Pang Wong, Nasrin Khazamipour, Soroush Alibagi, Joya Saade, Daria Golanarian, Negin Farivar, Mads Daugaard, Nader Al Nakouzi. Precise assessment of cancer cell growth and survival by artificial intelligence [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2311.

Label Noise Robust Crowd Counting with Loss Filtering Factor

ABSTRACT Crowd counting, a crucial computer vision task, aims at estimating the number of individuals in various environments. Each person in crowd counting datasets is typically annotated by a point at the center of the head. However, challenges like dense crowds, diverse scenarios, significant obscuration, and low resolution lead to inevitable label noise, adversely impacting model performance. Driven by the need to enhance model robustness in noisy environments and improve accuracy, we propose the Loss Filtering Factor (LFF) and the corresponding Label Noise Robust Crowd Counting (LNRCC) training scheme. LFF innovatively filters out losses caused by label noise during training, enabling models to focus on accurate data, thereby increasing reliability. Our extensive experiments demonstrate the effectiveness of LNRCC, which consistently improves performance across all models and datasets, with an average enhancement of 3.68% in Mean Absolute Error (MAE), 6.7% in Mean Squared Error (MSE) and 4.68% in Grid Average Mean Absolute Error (GAME). The universal applicability of this approach, coupled with its ease of integration into any neural network model architecture, marks a significant advancement in the field of computer vision, particularly in addressing the pivotal issue of accuracy in crowd counting under challenging conditions.

Flexible model-based non-negative matrix factorization with application to mutational signatures.

Somatic mutations in cancer can be viewed as a mixture distribution of several mutational signatures, which can be inferred using non-negative matrix factorization (NMF). Mutational signatures have previously been parametrized using either simple mono-nucleotide interaction models or general tri-nucleotide interaction models. We describe a flexible and novel framework for identifying biologically plausible parametrizations of mutational signatures, and in particular for estimating di-nucleotide interaction models. Our novel estimation procedure is based on the expectation-maximization (EM) algorithm and regression in the log-linear quasi-Poisson model. We show that di-nucleotide interaction signatures are statistically stable and sufficiently complex to fit the mutational patterns. Di-nucleotide interaction signatures often strike the right balance between appropriately fitting the data and avoiding over-fitting. They provide a better fit to data and are biologically more plausible than mono-nucleotide interaction signatures, and the parametrization is more stable than the parameter-rich tri-nucleotide interaction signatures. We illustrate our framework in a large simulation study where we compare to state of the art methods, and show results for three data sets of somatic mutation counts from patients with cancer in the breast, Liver and urinary tract.

The Zero-Inflated Poisson - Probit regression model: a new model for count data

This paper wishes to propose a new model for count data and it is shortly called a Zero-Inflated Poisson - Probit (ZIP-P) model. The way of setting up for this new model is also based on the traditional model, the Zero-Inflated Poisson (ZIP) model. Note that this new model as well as the ZIP model has a wide range of practical applications and it can be easily applied and analyzed to the real data sets. More specifically, both of these models can be applied very well to count data sets where the frequency of zeros occupies a very high percentage. To achieve the goal of this study, we first recall the construction and general formula of the ZIP model. Then we present the most detailed way for building a new model includes the general formula and related equations necessary for estimating the parameters of the ZIP-P model. In addition, asymptotic results, numerical simulations and analysis of a real data set are also performed in this work. Moreover, we also clarify that this actual data set fits the ZIP-P model better than the other models considered. Finally, several lines of discussions and inferences are made in the study.

ICNet: A Dual-Branch Instance Segmentation Network for High-Precision Pig Counting

A clear understanding of the number of pigs plays a crucial role in breeding management. Computer vision technology possesses several advantages, as it is harmless and labour-saving compared to traditional counting methods. Nevertheless, the existing methods still face some challenges, such as: (1) the lack of a substantial high-precision pig-counting dataset; (2) creating a dataset for instance segmentation can be time-consuming and labor-intensive; (3) interactive occlusion and overlapping always lead to incorrect recognition of pigs; (4) existing methods for counting such as object detection have limited accuracy. To address the issues of dataset scarcity and labor-intensive manual labeling, we make a semi-auto instance labeling tool (SAI) to help us to produce a high-precision pig counting dataset named Count1200 including 1220 images and 25,762 instances. The speed at which we make labels far exceeds the speed of manual annotation. A concise and efficient instance segmentation model built upon several novel modules, referred to as the Instances Counting Network (ICNet), is proposed in this paper for pig counting. ICNet is a dual-branch model ingeniously formed of a combination of several layers, which is named the Parallel Deformable Convolutions Layer (PDCL), which is trained from scratch and primarily composed of a couple of parallel deformable convolution blocks (PDCBs). We effectively leverage the characteristic of modeling long-range sequences to build our basic block and compute layer. Along with the benefits of a large effective receptive field, PDCL achieves a better performance for multi-scale objects. In the trade-off between computational resources and performance, ICNet demonstrates excellent performance and surpasses other models in Count1200, AP of 71.4% and AP50 of 95.7% are obtained in our experiments. This work provides inspiration for the rapid creation of high-precision datasets and proposes an accurate approach to pig counting.

PlantBiCNet: A new paradigm in plant science with bi-directional cascade neural network for detection and counting

Deep learning is increasingly popular for precise plant detection and counting in computer vision applications. Despite the rising cross-disciplinary research in this field, an important issue has been neglected: due to the inherent constraints imposed by the natural growth patterns of plants, the costly and time-consuming process of collecting plant image data, resulting in a scarcity of real-world available data. Consequently, researchers are often constrained to exploring large-scale data representations or limited applications. Currently, still no generalized network architecture excels in demanding environments. To address this, we introduce two new bi-directional cascade neural network paradigms, PlantBiCNet and its lightweight variant, PlantBiCNet-Lite. Unlike previous research, we place greater emphasis on the network decoding process itself rather than the encoder. Our bi-directional cascade and weight fusion decoding approach maximizes high-level semantics and low-level spatial data, enhancing the localization signal in a way that respects gradient diversity. We evaluate PlantBiCNet and PlantBiCNet-Lite on five public plant detection and counting datasets and a novel plant dataset. The average precision AP50 on these six small-scale datasets reached 0.905, with a high coefficient of determination R2 of 0.935, while the reported frame rate was 196 FPS. These results significantly outperform state-of-the-art computer vision methods. Moreover, we found that the introduction of the Dropout mechanism markedly enhances the model's performance and generalization ability. Further analysis supports our experiments and provides valuable insights into network optimization. Overall, our organization's open datasets and network modeling details provide a solid foundation for future research in the field of plant science.