Convergence Of Model Research Articles

Real Time Object Detection Algorithm in Foggy Weather Based on WVIT-YOLO Model

Introduction: To address the challenges of low visibility, object recognition difficulties, and low detection accuracy in foggy weather, this paper introduces the WViT-YOLO real-time fog detection model, built on the YOLOv5 framework. The NVIT-Net backbone network, incorporating NTB and NCB modules, enhances the model's ability to extract both global and local features from images. Method: An efficient convolutional C3_DSConv module is designed and integrated with channel attention mechanisms and ShuffleAttention at each upsampling stage, improving the model's computational speed and its ability to detect small and blurry objects. The Wise-IOU loss function is utilized during the prediction stage to enhance the model's convergence efficiency. Result: Experimental results on the publicly available RTTS dataset for vehicle detection in foggy conditions demonstrate that the WViT-YOLO model achieves a 3.2% increase in precision, a 9.5% rise in recall, and an 8.6% improvement in mAP50 compared to the baseline model. Furthermore, WViT-YOLO shows a 9.5% and 8.6% mAP50 improvement over YOLOv7 and YOLOv8, respectively. For detecting small and blurry objects in foggy conditions, the model demonstrates approximately a 5% improvement over the benchmark, significantly enhancing the detection network's generalization ability under foggy conditions. Conclusion: This advancement is crucial for improving vehicle safety in such weather. The code is available at https://github.com/QinghuaZhang1/mode.

Human motor learning dynamics in high-dimensional tasks.

Conventional approaches to enhance movement coordination, such as providing instructions and visual feedback, are often inadequate in complex motor tasks with multiple degrees of freedom (DoFs). To effectively address coordination deficits in such complex motor systems, it becomes imperative to develop interventions grounded in a model of human motor learning; however, modeling such learning processes is challenging due to the large DoFs. In this paper, we present a computational motor learning model that leverages the concept of motor synergies to extract low-dimensional learning representations in the high-dimensional motor space and the internal model theory of motor control to capture both fast and slow motor learning processes. We establish the model's convergence properties and validate it using data from a target capture game played by human participants. We study the influence of model parameters on several motor learning trade-offs such as speed-accuracy, exploration-exploitation, satisficing, and flexibility-performance, and show that the human motor learning system tunes these parameters to optimize learning and various output performance metrics.

ADWNet: An improved detector based on YOLOv8 for application in adverse weather for autonomous driving

AbstractDrawing inspiration from the state‐of‐the‐art object detection framework YOLOv8, a new model termed adverse weather net (ADWNet) is proposed. To enhance the model's feature extraction capabilities, the efficient multi‐scale attention (EMA) module has been integrated into the backbone. To address the problem of information loss in fused features, Neck has been replaced with RepGDNeck. Simultaneously, to expedite the model's convergence, the bounding box's loss function has been optimized to SIoU loss. To elucidate the advantages of ADWNet in the context of adverse weather conditions, ablation studies and comparative experiments were conducted. The results indicate that although the model's parameter count increased by 18.4%, the accuracy for detecting rain, snow, and fog in adverse weather conditions improved by 22%, while the FLOPs (floating point operations) decreased by 5%. The results of the comparison experiments conducted on the WEDGE dataset show that ADWNet outperforms other object detection models in adverse weather in terms of accuracy, model parameters and FLOPs. To validate ADWNet's real‐world efficacy, data was extracted from a car recorder under adverse conditions on highways, visual inference was conducted, and its accuracy was demonstrated in interpreting real‐world scenarios. The config files are available at https://github.com/Xinyun‐Feng/ADWNet.

Forest fire detection utilizing ghost Swin transformer with attention and auxiliary geometric loss

Forest fires are a devastating natural disaster. Existing fire detection models face limitations in dataset availability, multi-scale feature extraction, and locating obscured or small flames and smoke. To address these issues, we develop a dataset containing real and synthetic forest fire images, sourced from a UAV (Unmanned Aerial Vehicle) perspective. Additionally, we propose the Ghost Convolution Swin Transformer (GCST) module to extract multi-scale flame and smoke features from different receptive fields by integrating parallel Ghost convolution and Swin Transformer. Subsequently, we design a lightweight reparameterized rotation attention module, which captures interactions across channel and spatial dimensions to suppress background noise and focus on obscured flames and smoke. Finally, we introduce a loss function, called Efficient Auxiliary Geometric Intersection over Union (EAGIoU), which employs an auxiliary bounding box to accelerate the model's convergence while integrating the geometrical principles of the predicted and real bounding boxes to accurately locate small flames and smoke. Extensive experimental results demonstrate that our method achieves 75.2 % mAP@0.5 and 42 % mAP@0.5:0.95 with a frame rate of 239 frames per second, indicating a significant improvement in accuracy and real-time performance compared to state-of-the-art techniques. The code and datasets are available at https://github.com/luckylil/forest-fire-detection.

MCCM: multi-scale feature extraction network for disease classification and recognition of chili leaves.

Currently, foliar diseases of chili have significantly impacted both yield and quality. Despite effective advancements in deep learning techniques for the classification of chili leaf diseases, most existing classification models still face challenges in terms of accuracy and practical application in disease identification. Therefore, in this study, an optimized and enhanced convolutional neural network model named MCCM (MCSAM-ConvNeXt-MSFFM) is proposed by introducing ConvNeXt. The model incorporates a Multi-Scale Feature Fusion Module (MSFFM) aimed at better capturing disease features of various sizes and positions within the images. Moreover, adjustments are made to the positioning, activation functions, and normalization operations of the MSFFM module to further optimize the overall model. Additionally, a proposed Mixed Channel Spatial Attention Mechanism (MCSAM) strengthens the correlation between non-local channels and spatial features, enhancing the model's extraction of fundamental characteristics of chili leaf diseases. During the training process, pre-trained weights are obtained from the Plant Village dataset using transfer learning to accelerate the model's convergence. Regarding model evaluation, the MCCM model is compared with existing CNN models (Vgg16, ResNet34, GoogLeNet, MobileNetV2, ShuffleNet, EfficientNetV2, ConvNeXt), and Swin-Transformer. The results demonstrate that the MCCM model achieves average improvements of 3.38%, 2.62%, 2.48%, and 2.53% in accuracy, precision, recall, and F1 score, respectively. Particularly noteworthy is that compared to the original ConvNeXt model, the MCCM model exhibits significant enhancements across all performance metrics. Furthermore, classification experiments conducted on rice and maize disease datasets showcase the MCCM model's strong generalization performance. Finally, in terms of application, a chili leaf disease classification website is successfully developed using the Flask framework. This website accurately identifies uploaded chili leaf disease images, demonstrating the practical utility of the model.

Dynamic economic dispatch of integrated energy system based on generative adversarial imitation learning

Within the context of energy structure transformation, the inherent uncertainty of new energy sources presents severe challenges to the optimized operation of integrated energy system. Model-driven optimization methods are limited by the accuracy and complexity of models, resulting in lower solving efficiency. Meanwhile, deep reinforcement learning methods that have gained attention in recent years face difficulties in designing rewards and slow convergence due to dealing with high-dimensional and complex dynamic information of integrated energy system. Therefore, inspired by imitation learning, this paper proposes a real-time dispatching method for integrated energy system based on generative adversarial imitation learning. Firstly, a generator network model is designed within the deep reinforcement learning framework, and expert prior knowledge is introduced to assist the generator network in iterative learning, enhancing the model's convergence effect. Secondly, drawing on the generative adversarial concept, a discriminator network model is developed to recognize expert and generated strategies, further designing game rewards to assist in updating the generator network parameters and avoiding the impact of subjectively defined reward functions on dispatching decisions. Finally, simulation analysis demonstrates that the proposed method can calculate the operation optimization solutions for integrated energy system more efficiently, significantly improving the accuracy of dispatching decisions and the model's convergence efficiency.

CFL：Data distribution edge clustering algorithm based on deep Federated learning

Federated learning (FL) allows end-devices to train local models on their respective local datasets and collaborate with the server to train a global predictive model, thus achieving the goal of machine learning while protecting the privacy and sensitive data of end-devices. However, simultaneous access to the server by a large number of end devices may result in increased transmission latency and some local models may have malicious behavior, converging in the opposite direction to the global model. As a result, additional communication costs can occur during the federation learning process. Existing research has mainly focused on reducing the number of communication rounds or cleaning dirty local data. In order to decrease the overall amount of local updates, we provide an edge-based model cleaning and device clustering strategy in this study. By computing the cosine similarity between local update parameters and global model parameters over a multi-dimensional space, the approach assesses if local updates are necessary while preventing pointless communications. In addition, end devices are grouped together based on where they are connected to the network and connect to the cloud via mobile edge nodes in clusters, therefore lowering the latency associated with high concurrent server access. Convolutional neural networks and Soft max regression are used, for instance, to perform MNIST handwritten digit recognition, and the effectiveness of the suggested approach to enhancing communication efficiency is confirmed. According to experimental findings, the edge-based model cleaning and device clustering technique decreases the quantity of local updates by 60% and speeds up the model's convergence by 10.3% when compared to the conventional FL approach.

Performance Evaluation of VGG_19 Model for Brain Tumor Datasets

This paper presents a comprehensive assessment of the VGG19 model for detecting brain tumors using deep learning methods, particularly convolutional neural networks (CNNs). The study evaluates key performance metrics such as accuracy, precision, recall, F1 score, and area under the curve (AUC). The investigation delves into the influence of different activation functions, learning rates, and epochs on the VGG19 model's performance. Activation functions are crucial for feature extraction and model convergence, with a comparison made between ReLU and Leaky ReLU to determine their effectiveness in enhancing brain tumor detection accuracy. Additionally, the study examines the impact of various learning rates (0.01, 0.001, 0.0001) on the model's convergence speed and overall performance. Different numbers of training epochs (15, 25, 50) are also considered to strike a balance between computational efficiency and model effectiveness. Evaluation metrics encompass accuracy, precision, recall, F1 score, and AUC, providing a comprehensive assessment of the model's performance across different configurations. The goal is to identify optimal combinations of activation functions, learning rates, and epochs to maximize the VGG19 model's accuracy in detecting brain tumors. This research contributes to advancing efficient and accurate methodologies for brain tumor detection, facilitating early diagnosis and treatment planning. The insights gained from this study can guide the development of more effective deep learning-based systems for medical image analysis, benefiting both patients and healthcare professionals.

Rapid Strawberry Ripeness Detection And 3D Localization of Picking Point Based on Improved YOLO V8-Pose with RGB-Camera

Accurate identification of strawberries at different growth stages as well as determination of optimal picking points by strawberry picking robots is a key issue in the field of agricultural automation. In this paper, a fast detection method of strawberry ripeness and picking point based on improved YOLO V8-Pose (You Only Look Once) and RGB-D depth camera is proposed to address this problem. By comparing the YOLO v5-Pose, YOLO v7-Pose, and YOLO v8-Pose models, it is determined to use the YOLO v8-Pose model as the fundamental model for strawberry ripeness and picking point detection. For the sake of further improving the accuracy of the model detection, this paper makes targeted improvements: all the Concat modules at the Neck part are replaced with BiFPN richer feature fusion, which enhances the global feature extraction capability of the model; the MobileViTv3 framework is employed to restructure the backbone network, thereby augmenting the model's capacity for contextual feature extraction. Subsequently, the output-side CIoU loss function is supplanted with the SIoU loss function, leading to an acceleration in the model's convergence. The enhanced YOLO v8-Pose demonstrates a 97.85% mAP-kp value, reflecting a 5.49% improvement over the initial model configuration.. For the sake of accurately localizing the three-dimensional information of strawberry picking points, the strawberry picking points are further projected into the corresponding depth information to obtain their three-dimensional information. The experimental results show that the mean absolute error and the mean absolute percentage error of strawberry picking point localization in this paper are 0.63 cm and 1.16%, respectively. In this study, we introduce a method capable of concurrently detecting strawberry maturity and identifying the precise harvesting location while accurately localizing the picking point. This investigation holds considerable theoretical and pragmatic relevance in augmenting the intelligence of strawberry harvesting robots and actualizing automation and smart capabilities in agricultural production.

Enhancing Spam Detection: A Crow-Optimized FFNN with LSTM for Email Security

Email security is paramount in today's digital landscape, as the proliferation of spam emails poses a significant threat to individuals and organizations alike. To combat this menace, this study introduces a novel approach that marries the power of Crow Search Optimization (CSO) with a Feedforward Neural Network (FFNN) and Long Short-Term Memory (LSTM) architecture to bolster spam detection. The proposed Crow-Optimized FFNN with LSTM (C-FFNN-LSTM) leverages CSO to fine-tune the neural network's parameters, optimizing its ability to distinguish between legitimate emails and spam. The CSO algorithm mimics the collaborative behavior of crows, thereby enhancing the model's convergence and robustness. Experimental results showcase the effectiveness of the C-FFNN-LSTM approach, achieving remarkable accuracy rates and reducing false positives. This innovation not only enhances email security but also offers a promising avenue for refining spam detection algorithms across various domains. In an era of ever-evolving cyber threats, the C-FFNN-LSTM framework stands as a beacon of improved email security, safeguarding digital communication channels.
 In our methodology, we attained an outstanding accuracy level of 99.1% during the testing phase.

AT4CTR: Auxiliary Match Tasks for Enhancing Click-Through Rate Prediction

Click-through rate (CTR) prediction is a vital task in industrial recommendation systems. Most existing methods focus on the network architecture design of the CTR model for better accuracy and suffer from the data sparsity problem. Especially in industrial recommendation systems, the widely applied negative sample down-sampling technique due to resource limitation worsens the problem, resulting in a decline in performance. In this paper, we propose Auxiliary Match Tasks for enhancing Click-Through Rate (AT4CTR) prediction accuracy by alleviating the data sparsity problem. Specifically, we design two match tasks inspired by collaborative filtering to enhance the relevance modeling between user and item. As the "click" action is a strong signal which indicates the user's preference towards the item directly, we make the first match task aim at pulling closer the representation between the user and the item regarding the positive samples. Since the user's past click behaviors can also be treated as the user him/herself, we apply the next item prediction as the second match task. For both the match tasks, we choose the InfoNCE as their loss function. The two match tasks can provide meaningful training signals to speed up the model's convergence and alleviate the data sparsity. We conduct extensive experiments on one public dataset and one large-scale industrial recommendation dataset. The result demonstrates the effectiveness of the proposed auxiliary match tasks. AT4CTR has been deployed in the real industrial advertising system and has gained remarkable revenue.

A combination prediction model based on Theil coefficient and induced continuous aggregation operator for the prediction of Shanghai composite index

This paper proposes an interval combination prediction model for Shanghai composite index, utilizing the Theil coefficient and the induced continuous generalized ordered weighted logarithmic harmonic averaging (ICGOWLHA) operator. The effectiveness of the proposed model under specific weight conditions and the existence of its analytical solution are demonstrated. Shanghai composite index's case analysis demonstrates that, in terms of interval root mean squared error (IRMSE), interval mean absolute error (IMAE), interval mean absolute percentage error (IMAPE), and interval mean squared percentage error (IMSPE), the proposed model's predictive performance improvements over the best-performing single prediction model are 29.33%, 25.72%, 26.10%, and 28.86%, respectively. At the same time, the theoretical properties of the model are verified in the results of the case analysis, and the model's convergence is reflected in sensitivity analysis. Through extensive model comparisons, it is observed that the model proposed in this paper exhibits strong generalization, without specific limitations on data size or feature count. It demonstrates good aggregation prediction performance for interval data. Moreover, it is applicable to various fields, including finance, environment, and others.

CCDN-DETR: A Detection Transformer Based on Constrained Contrast Denoising for Multi-Class Synthetic Aperture Radar Object Detection.

The effectiveness of the SAR object detection technique based on Convolutional Neural Networks (CNNs) has been widely proven, and it is increasingly used in the recognition of ship targets. Recently, efforts have been made to integrate transformer structures into SAR detectors to achieve improved target localization. However, existing methods rarely design the transformer itself as a detector, failing to fully leverage the long-range modeling advantages of self-attention. Furthermore, there has been limited research into multi-class SAR target detection. To address these limitations, this study proposes a SAR detector named CCDN-DETR, which builds upon the framework of the detection transformer (DETR). To adapt to the multiscale characteristics of SAR data, cross-scale encoders were introduced to facilitate comprehensive information modeling and fusion across different scales. Simultaneously, we optimized the query selection scheme for the input decoder layers, employing IOU loss to assist in initializing object queries more effectively. Additionally, we introduced constrained contrastive denoising training at the decoder layers to enhance the model's convergence speed and improve the detection of different categories of SAR targets. In the benchmark evaluation on a joint dataset composed of SSDD, HRSID, and SAR-AIRcraft datasets, CCDN-DETR achieves a mean Average Precision (mAP) of 91.9%. Furthermore, it demonstrates significant competitiveness with 83.7% mAP on the multi-class MSAR dataset compared to CNN-based models.

Identification of cotton pest and disease based on CFNet- VoV-GCSP -LSKNet-YOLOv8s: a new era of precision agriculture.

The study addresses challenges in detecting cotton leaf pests and diseases under natural conditions. Traditional methods face difficulties in this context, highlighting the need for improved identification techniques. The proposed method involves a new model named CFNet-VoV-GCSP-LSKNet-YOLOv8s. This model is an enhancement of YOLOv8s and includes several key modifications: (1) CFNet Module. Replaces all C2F modules in the backbone network to improve multi-scale object feature fusion. (2) VoV-GCSP Module. Replaces C2F modules in the YOLOv8s head, balancing model accuracy with reduced computational load. (3) LSKNet Attention Mechanism. Integrated into the small object layers of both the backbone and head to enhance detection of small objects. (4) XIoU Loss Function. Introduced to improve the model's convergence performance. The proposed method achieves high performance metrics: Precision (P), 89.9%. Recall Rate (R), 90.7%. Mean Average Precision (mAP@0.5), 93.7%. The model has a memory footprint of 23.3MB and a detection time of 8.01ms. When compared with other models like YOLO v5s, YOLOX, YOLO v7, Faster R-CNN, YOLOv8n, YOLOv7-tiny, CenterNet, EfficientDet, and YOLOv8s, it shows an average accuracy improvement ranging from 1.2% to 21.8%. The study demonstrates that the CFNet-VoV-GCSP-LSKNet-YOLOv8s model can effectively identify cotton pests and diseases in complex environments. This method provides a valuable technical resource for the identification and control of cotton pests and diseases, indicating significant improvements over existing methods.

Improved RepVGG ground-based cloud image classification with attention convolution

Abstract. Atmospheric clouds greatly impact Earth's radiation, hydrological cycle, and climate change. Accurate automatic recognition of cloud shape based on a ground-based cloud image is helpful for analyzing solar irradiance, water vapor content, and atmospheric motion and then predicting photovoltaic power, weather trends, and severe weather changes. However, the appearance of clouds is changeable and diverse, and their classification is still challenging. In recent years, convolution neural networks (CNNs) have made great progress in ground-based cloud image classification. However, traditional CNNs poorly associate long-distance clouds, making the extraction of global features of cloud images quite problematic. This study attempts to mitigate this problem by elaborating on a ground-based cloud image classification method based on the improved RepVGG convolution neural network and attention mechanism. Firstly, the proposed method increases the RepVGG residual branch and obtains more local detail features of cloud images through small convolution kernels. Secondly, an improved channel attention module is embedded after the residual branch fusion, effectively extracting the global features of cloud images. Finally, the linear classifier is used to classify the ground cloud images. Finally, the warm-up method is applied to optimize the learning rate in the training stage of the proposed method, making it lightweight in the inference stage and thus avoiding overfitting and accelerating the model's convergence. The proposed method is validated on the multimodal ground-based cloud dataset (MGCD) and the ground-based remote sensing cloud database (GRSCD) containing seven cloud categories, with the respective classification accuracy rate values of 98.15 % and 98.07 % outperforming those of the 10 most advanced methods used as the reference. The results obtained are considered instrumental in ground-based cloud image classification.

Efficient click-based interactive segmentation for medical image with improved Plain-ViT.

The primary objective of interactive medical image segmentation systems is to achieve more precise segmentation outcomes with reduced human intervention. This endeavor holds significant clinical importance for both pre-diagnostic pathological assessments and prognostic recovery. Among the various interaction methods available, click-based interactions stand out as an intuitive and straightforward approach compared to alternatives such as graffiti, bounding boxes, and extreme points. To improve the model's ability to interpret click-based interactions, we propose a comprehensive interactive segmentation framework that leverages an iterative weighted loss function based on user clicks. To enhance the segmentation capabilities of the Plain-ViT backbone, we introduce a Residual Multi-Headed Self-Attention encoder with hierarchical inputs and residual connections, offering multiple perspectives on the data. This innovative architecture leads to a remarkable improvement in segmentation model performance. In this research paper, we assess the robustness of our proposed framework using a self-compiled T2-MRI image dataset of the prostate and three publicly available datasets containing images of other organs. Our experimental results convincingly demonstrate that our segmentation model surpasses existing state-of-the-art methods. Furthermore, the incorporation of an iterative loss function training strategy significantly accelerates the model's convergence rate during interactions. In the prostate dataset, we achieved an impressive Intersection over Union (IoU) score of 88.11% and Number of Clicks(NoC) at 80% are 7.03 clicks.

A new mathematical formulation for the simultaneous optimization of charging infrastructure and vehicle schedules for electric bus systems

This paper presents a new mixed integer problem (MIP) formulation for the Charging Location and Electric Vehicle Scheduling Problem (CLEVSP). The CLEVSP aims at minimizing the investment costs for charging infrastructure and electric buses as well as the operational costs resulting from the vehicle schedule of an electric bus system applying opportunity charging. Leveraging the assumption of a homogeneous vehicle fleet, the presented MIP is based on two-index decision variables for vehicle scheduling. The formulation allows finding near-optimal solutions for real-world instances of small and medium-sized cities and estimating their solution quality. The computational experiments show that the model, when solved with the Gurobi optimization software, is competitive with state-of-the-art solution methods for the CLEVSP even for problem instances with several thousand service trips. The model's convergence behavior strongly depends on the difficulty of the problem, which is not only influenced by the number of service trips but also by the available battery capacity of the electric buses.

Optimization and control in rubella transmission dynamics: A boundedness-preserving numerical model with vaccination

Rubella is an infectious disease that can spread globally. It spreads in the subtropics as well as the tropics. Although it is commonly thought to be a non-fatal condition, there are several circumstances in which it can be fatal. Pregnant women infected with the Rubella virus face a high risk of fetal development. The main goal of this study is to look into a model for the spread of Rubella while considering a vaccination campaign as a control measure. The positivity and boundedness of the re-infection rubella transmission and vaccination model are investigated. The local and global stability of the equilibrium points is examined. The sensitivity of parameters is also investigated. Three different techniques forward Euler, RK-4, and non-standard finite difference (NSFD) method are developed for the numerical solution, and their simulation results are examined. Among these three, the NSFD method is superior due to its convergence and positive behavior for all step size values. For some values of the step size, the other two techniques failed to produce positivity and convergent solutions. Analytically, the proposed model's convergence, positivity, boundedness, and consistency are investigated. Finally, the impact of the Rubella vaccine on infected populations has been examined, revealing that vaccination is one of the most effective ways to prevent rubella transmission.

A Precise Framework for Rice Leaf Disease Image-Text Retrieval Using FHTW-Net.

Cross-modal retrieval for rice leaf diseases is crucial for prevention, providing agricultural experts with data-driven decision support to address disease threats and safeguard rice production. To overcome the limitations of current crop leaf disease retrieval frameworks, we focused on four common rice leaf diseases and established the first cross-modal rice leaf disease retrieval dataset (CRLDRD). We introduced cross-modal retrieval to the domain of rice leaf disease retrieval and introduced FHTW-Net, a framework for rice leaf disease image-text retrieval. To address the challenge of matching diverse image categories with complex text descriptions during the retrieval process, we initially employed ViT and BERT to extract fine-grained image and text feature sequences enriched with contextual information. Subsequently, two-way mixed self-attention (TMS) was introduced to enhance both image and text feature sequences, with the aim of uncovering important semantic information in both modalities. Then, we developed false-negative elimination-hard negative mining (FNE-HNM) strategy to facilitate in-depth exploration of semantic connections between different modalities. This strategy aids in selecting challenging negative samples for elimination to constrain the model within the triplet loss function. Finally, we introduced warm-up bat algorithm (WBA) for learning rate optimization, which improves the model's convergence speed and accuracy. Experimental results demonstrated that FHTW-Net outperforms state-of-the-art models. In image-to-text retrieval, it achieved R@1, R@5, and R@10 accuracies of 83.5%, 92%, and 94%, respectively, while in text-to-image retrieval, it achieved accuracies of 82.5%, 98%, and 98.5%, respectively. FHTW-Net offers advanced technical support and algorithmic guidance for cross-modal retrieval of rice leaf diseases.

Galaxy recognition based on improved efficientNetV2S

In order to identify effective metrics that can accurately duplicate the probability distributions resulting from human classifications, this paper analyzes an improved approach for galaxy morphologies classification. At the present stage, this field still faces the problem of insufficient quality and quantity of image data, low accuracy of computer recognition and weak generalization ability of the model. From the previous research, Convolution Neural Network (CNN) can be a valid technique to complete this task but usually spends a large time and space complexity. For the purpose of increasing effectiveness, this paper improves EfficientNetV2S to construct recognition models and characterize their performance in galaxy recognition. The procedure includes data preparation and augmentation, model structure creation, attention mechanism addition, fine-tuning, and result visualization. A Fused mobile inverted bottleneck convolution (Fuse-MBConv) structure was used to accelerate the model's convergence speed. Besides, the Convolutional block attention module (CBAM) was used to improve performance and feature representation capabilities. The model in this study can minimize complexity with the number of parameters and utilize less memory while maintaining excellent accuracy. This research is conducted on the Galaxy10 DECals dataset. Experimental results show that it achieves an 87% high precision with 20.6m parameters which is more efficient than models currently used in previous research.