Identification Dataset Research Articles

AFPILD: Acoustic footstep dataset collected using one microphone array and LiDAR sensor for person identification and localization

Empowering the auditory perception-based person identification and localization could enable the service robots to navigate among the crowds in a more socially compliant manner. And it is more user-friendly with fewer privacy concerns compared to the widely studied camera-supported approaches. One way to achieve such auditory person identification and localization is to identify and locate the subject by hearing her/his footstep. As far as we know, although several related acoustic identification benchmark datasets have been made, the dataset that simultaneously concentrates on acoustic identification and localization is absent. Aiming at the shortage of such benchmark datasets, we propose to build the acoustic footstep-based person identification and localization dataset (AFPILD) by unifying the identify and locate tasks for the first time, concerning the clothing and shoe type covariates from the human-aware mobile service robot navigation perspective. Specifically, we record the walking sounds of 40 subjects with a lightweight microphone array in one real unstrained indoor room. For ground truth localization label annotation, we record the point clouds of the walking subjects with a multi-line LiDAR synchronously. The AFPILD contains nearly 10 h of footstep audios (around 62,174 footstep events). To the best of our knowledge, this is the largest and most comprehensive conducted acoustic footstep dataset built for person identification and localization study up to now. We also present the baseline works concerning the identification, localization, and synchronous identify-locate tasks, respectively, to compare various potential approaches on this dataset in the future. Experimental results show that the covariates of clothing and shoe type degrade the footstep identification (around 12% of clothing and 33% of shoe type), while their impacts on the localization are unobvious.

Local binary pattern and its derivatives to handwriting-based gender classification

Several studies by psychologists and computer scientists have verified the link between handwriting and writer gender. The texture of the writing image is a major indicator of whether it is male or female writing. This paper conducts a comparison analysis to examine the effectiveness of various local binary patterns (LBPs) techniques in detecting gender from scanned images of handwriting. We study different LBP variants, including complete local binary pattern (CLBP), local ternary pattern (LTP), local configuration pattern (LCP), rotated local binary pattern (RLBP), local binary pattern variance (LBPV), and multi-scale local binary pattern (MLBP), as features for representing handwriting images. A support vector machine (SVM) is trained using features from male and female writing. The method achieves encouraging classification rates of 76.68 when tested on subsets of the Qatar University writer identification (QUWI) dataset containing English and Arabic writing samples when using the experimental protocols of the International Conference on Document Analysis and Recognition (ICDAR) 2013 gender classification competitions.

Interpretable Incremental Voltage–Current Representation Attention Convolution Neural Network for Nonintrusive Load Monitoring

This paper proposes an interpretable incremental voltage-current representation attention convolution neural network for the non-intrusive load monitoring (NILM) task. The proposed method consists of two parts: (i) the voltage-current representation attention mechanism in the proposed network is designed in collaboration with the data pre-processing method. They provide the role for the classification function of neural networks.; (ii) this paper proposed an adaptive distillation incremental learning method that introduced incremental learning into the NILM field. In this work, the public dataset plug-load appliance identification dataset is used to validate the proposed voltage-current representation attention mechanism and adaptive distillation incremental learning method in this paper. In addition, the performance of the proposed algorithms is also complemented in this paper using a private dataset. According to the experimental results, the performance of the proposed method in this paper is better than the comparison methods.

Who is Alyx? A new behavioral biometric dataset for user identification in XR

Introduction: This paper addresses the need for reliable user identification in Extended Reality (XR), focusing on the scarcity of public datasets in this area.Methods: We present a new dataset collected from 71 users who played the game “Half-Life: Alyx” on an HTC Vive Pro for 45 min across two separate sessions. The dataset includes motion and eye-tracking data, along with physiological data from a subset of 31 users. Benchmark performance is established using two state-of-the-art deep learning architectures, Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU).Results: The best model achieved a mean accuracy of 95% for user identification within 2 min when trained on the first session and tested on the second.Discussion: The dataset is freely available and serves as a resource for future research in XR user identification, thereby addressing a significant gap in the field. Its release aims to facilitate advancements in user identification methods and promote reproducibility in XR research.

Determining the parent and associated fragment formulae in mass spectrometry via the parent subformula graph

BackgroundIdentifying the molecular formula and fragmentation reactions of an unknown compound from its mass spectrum is crucial in areas such as natural product chemistry and metabolomics. We propose a method for identifying the correct candidate formula of an unidentified natural product from its mass spectrum. The method involves scoring the plausibility of parent candidate formulae based on a parent subformula graph (PSG), and two possible metrics relating to the number of edges in the PSG. This method is applicable to both electron-impact mass spectrometry (EI-MS) and tandem mass spectrometry (MS/MS) data. Additionally, this work introduces the two-dimensional fragmentation plot (2DFP) for visualizing PSGs.ResultsOur results suggest that incorporating information regarding the edges of the PSG results in enhanced performance in correctly identifying parent formulae, in comparison to the more well-accepted “MS/MS score”, on the 2016 Computational Assessment of Small Molecule Identification (CASMI 2016) data set (76.3 vs 58.9% correct formula identification) and the Research Centre for Toxic Compounds in the Environment (RECETOX) data set (66.2% vs 59.4% correct formula identification). In the extension of our method to identify the correct candidate formula from complex EI-MS data of semiochemicals, our method again performed better (correct formula appearing in the top 4 candidates in 20/23 vs 7/23 cases) than the MS/MS score, and enables the rapid identification of both the correct parent ion mass and the correct parent formula with minimal expert intervention.ConclusionOur method reliably identifies the correct parent formula even when the mass information is ambiguous. Furthermore, should parent formula identification be successful, the majority of associated fragment formulae can also be correctly identified. Our method can also identify the parent ion and its associated fragments in EI-MS spectra where the identity of the parent ion is unclear due to low quantities and overlapping compounds. Finally, our method does not inherently require empirical fitting of parameters or statistical learning, meaning it is easy to implement and extend upon.Scientific contributionDeveloped, implemented and tested new metrics for assessing plausibility of candidate molecular formulae obtained from HR-MS data.

MammalClub: An Annotated Wild Mammal Dataset for Species Recognition, Individual Identification, and Behavior Recognition

Mammals play an important role in conserving species diversity and maintaining ecological balance, so research on mammal species composition, individual identification, and behavioral analysis is of great significance for optimizing the ecological environment. Due to their great capabilities for feature extraction, deep learning networks have gradually been applied to wildlife monitoring. However, training a network requires a large number of animal image samples. Although a few wildlife datasets contain many mammals, most mammal images in these datasets are not annotated. In particular, selecting mammalian images from vast and comprehensive datasets is still a time-consuming task. Therefore, there is currently a lack of specialized datasets of images of wild mammals. To address these limitations, this article created a mammal image dataset (named MammalClub), which contains three sub-datasets (i.e., a species recognition sub-dataset, an individual identification sub-dataset, and a behavior recognition sub-dataset). This study labeled the bounding boxes of the images used for species recognition and the coordinates of the mammals’ skeletal joints for behavior recognition. This study also captured images of each individual from different points of view for individual mammal identification. This study explored novel intelligent animal recognition models and compared and analyzed them with the mainstream models in order to test the dataset.

MMBERT: a unified framework for biomedical named entity recognition.

Named entity recognition (NER) is an important task in natural language processing (NLP). In recent years, NER has attracted much attention in the biomedical field. However, due to the lack of biomedical named entity identification datasets, the complexity and rarity of biomedical named entities and so on, biomedical NER is more difficult than general domain NER. So in this paper, we propose a framework (MMBERT) based on Transformer to solve the problems above. To address the scarcity of biomedical named entity recognition datasets, we introduce ERNIE-Health, a new Chinese language representation model pre-trained on large-scale biomedical text corpora. Because of the complexity and rarity of biomedical named entities, we use the Bert and CW-LSTM structures to get the joint feature vector of word pairs relations. In addition, we design multi-granularity 2D convolution to refine the relationship and representation between word pairs. Finally, we design a convolutional neural network (CNN) structure and a co-predictor to improve the model's generalization capability and prediction accuracy. We have conducted extensive experiments on three benchmark datasets, and the experimental results show that our model achieves the best results compared with several baseline models in the experiment.

ClinicalOmicsDB: exploring molecular associations of oncology drug responses in clinical trials.

Matching patients to optimal treatment is challenging, in part due to the limited availability of real-world clinical datasets for predictive biomarker identification. The growing integration of omics profiling into clinical trials presents a new opportunity to tackle this challenge. Here, we introduce ClinicalOmicsDB, a web application for exploring molecular associations of oncology drug responses in clinical trials. This database includes transcriptomic data from 40 clinical trial studies, with 5913 patients spanning 11 cancer types. These studies include 67 treatment arms with a variety of chemotherapy, targeted therapy and immunotherapy drugs, and their combinations, which we organize based on an established ontology for easier navigation. The web application provides users with three options to explore molecular associations of oncology drug responses, focusing on studies, treatments or genes, respectively. Gene set analysis further connects treatment response to pathway activity and tumor microenvironment attributes. The user-friendly web interface of ClinicalOmicsDB streamlines interactive analysis. A Rust-based backend speeds up response time, and application programming interfaces and an R package enable programmatic access. We use three case studies to demonstrate the utility of this resource in human cancer studies. ClinicalOmicsDB is freely available at http://trials.linkedomics.org/.

Classifying cuneiform symbols using machine learning algorithms with unigram features on a balanced dataset

Abstract Problem Recognizing written languages using symbols written in cuneiform is a tough endeavor due to the lack of information and the challenge of the process of tokenization. The Cuneiform Language Identification (CLI) dataset attempts to understand seven cuneiform languages and dialects, including Sumerian and six dialects of the Akkadian language: Old Babylonian, Middle Babylonian Peripheral, Standard Babylonian, Neo-Babylonian, Late Babylonian, and Neo-Assyrian. However, this dataset suffers from the problem of imbalanced categories. Aim Therefore, this article aims to build a system capable of distinguishing between several cuneiform languages and solving the problem of unbalanced categories in the CLI dataset. Methods Oversampling technique was used to balance the dataset, and the performance of machine learning algorithms such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), and deep learning such as deep neural networks (DNNs) using the unigram feature extraction method was investigated. Results The proposed method using machine learning algorithms (SVM, KNN, DT, and RF) on a balanced dataset obtained an accuracy of 88.15, 88.14, 94.13, and 95.46%, respectively, while the DNN model got an accuracy of 93%. This proves improved performance compared to related works. Conclusion This proves the improvement of classifiers when working on a balanced dataset. The use of unigram features also showed an improvement in the performance of the classifier as it reduced the size of the data and accelerated the processing process.

Deep dissection of stemness-related hierarchies in hepatocellular carcinoma

BackgroundIncreasing evidence suggests that hepatocellular carcinoma (HCC) stem cells (LCSCs) play an essential part in HCC recurrence, metastasis, and chemotherapy and radiotherapy resistance. Multiple studies have demonstrated that stemness-related genes facilitate the progression of tumors. However, the mechanism by which stemness-related genes contribute to HCC is not well understood. Here, we aim to construct a stemness-related score (SRscores) model for deeper analysis of stemness-related genes, assisting with the prognosis and individualized treatment of HCC patients.Further, we found that the gene LPCAT1 was highly expressed in tumor tissues by immunohistochemistry, and sphere-forming assay revealed that knockdown of LPCAT1 inhibited the sphere-forming ability of hepatocellular carcinoma cells.MethodsWe used the TCGA-LIHC dataset to screen stemness-related genes of HCC from the MSigDB database. Prognosis, tumor microenvironment, immunological checkpoints, tumor immune dysfunction, rejection, treatment sensitivity, and putative biological pathways were examined. Random forest created the SRscores model. The anti-PD-1/anti-CTLA4 immunotherapy, tumor mutational burden, medication sensitivity, and cancer stem cell index were compared between the high- and low-risk score groups. We also examined risk scores for different cell types using single-cell RNA sequencing data and correlated transcription factor activity in cancer stem cells with SRscores genes. Finally, we tested core marker expression and biological functions.ResultsPatients can be divided into two subtypes (Cluster1 and Cluster2) based on the TCGA-LIHC dataset's identification of 11 stemness-related genes. Additionally, a SRscores was developed based on subtypes. Cluster2 and the group with the lowest SRscores had superior survival and immunotherapy response than Cluster1 and the group with the highest SRscores. The group with a high SRscores was significantly more enriched in classical tumor pathways than the group with a low SRscores. Multiple transcription factors and SRscores genes are correlated. The core gene LPCAT1 is highly expressed in rat liver cancer tissues and promotes tumor cell sphere formation.ConclusionA SRscores model can be utilized to predict the prognosis of HCC patients as well as their response to immunotherapy.

A Contrast-Enhanced CT-Based Deep Learning System for Preoperative Prediction of Colorectal Cancer Staging and RAS Mutation.

This study aimed to build a deep learning system using enhanced computed tomography (CT) portal-phase images for predicting colorectal cancer patients' preoperative staging and RAS gene mutation status. The contrast-enhanced CT image dataset comprises the CT portal-phase images from a retrospective cohort of 231 colorectal cancer patients. The deep learning system was developed via migration learning for colorectal cancer detection, staging, and RAS gene mutation status prediction. This study used pre-trained Yolov7, vision transformer (VIT), swin transformer (SWT), EfficientNetV2, and ConvNeXt. 4620, and contrast-enhanced CT images and annotated tumor bounding boxes were included in the tumor identification and staging dataset. A total of 19,700 contrast-enhanced CT images comprise the RAS gene mutation status prediction dataset. In the validation cohort, the Yolov7-based detection model detected and staged tumors with a mean accuracy precision (IoU = 0.5) (mAP_0.5) of 0.98. The area under the receiver operating characteristic curve (AUC) in the test set and validation set for the VIT-based prediction model in predicting the mutation status of the RAS genes was 0.9591 and 0.9554, respectively. The detection network and prediction network of the deep learning system demonstrated great performance in explaining contrast-enhanced CT images. In this study, a deep learning system was created based on the foundation of contrast-enhanced CT portal-phase imaging to preoperatively predict the stage and RAS mutation status of colorectal cancer patients. This system will help clinicians choose the best treatment option to increase colorectal cancer patients' chances of survival and quality of life.

An IMU dataset for human thermal comfort activities identification: Experimental designs and applications

Thermal comfort of occupants is key feedback information for improving indoor environment and managing building energy use. Through analyzing inertial measurement units (IMU) data from wearable devices with machine learning, thermal comfort of occupants can be detected in a non- intrusive method. This paper proposed a dataset consisted of IMU data collected from 30 participants (14 males and 16 females, aged 23.23 ± 1.70 years, height 168.67 ± 8.02 cm, and weight 59.55 ± 10.96 kg) who wore two IMUs on their hands while performing 30 thermal comfort activities (10 cold-related, 10 hot-related, and 10 neutral activities) according to their personal habits.The database is divided into two parts: (1) Single activities data, which includes 4500 samples acquired from experiments where each participant was asked to perform 30 thermal comfort activities individually. (2) Continuous multi-activity data, which comprise 360 samples collected while participants performed a series of randomly assigned activities in a more natural and continuous manner. The combination of these two parts provides a comprehensive dataset for both the training and testing phases of machine learning models. By offering detailed labels, this database aims to serve as a foundation for research exploring machine learning applications in detecting occupant thermal comfort, ultimately contributing to improved indoor environments and more efficient building energy management.

Intelligent vision for the detection of chemistry glassware toward AI robotic chemists

One of the key steps to make an artificially intelligent (AI) and robotic chemist is the introduction of machine vision for guiding the experiment operation in the AI-redefined laboratory. In order to realize the targets, the prerequisites are to innovate/implement the intelligent vision for the detection of chemistry glassware. Here, we reported a computer vision method based on You only look once (YOLO) with a self-developed Chemical Vessel Identification Dataset (CViG) for the improvement of classification and recognition performance. The training dataset has been collected that includes 4072 images in real-time chemical laboratory. Three models, YOLOv5s, Slim-YOLOv5s and YOLOv7, have been exploited for the recognition of seven types of glassware in the condition of different scenarios (recognition distance, light and dark, stationary and moving). The improved Slim-YOLOv5s exhibited better recognition ability in various scenes, and the recognition accuracy of chemical vessels is improved by 1.51 % compared with YOLOv5s, and the size of the model is reduced from 14.4 MB to 11.0 MB. Slim-YOLOv5s's mAP is similar to YOLOv7's ability with a disadvantage of large volume, suggested that the improved Slim-YOLOv5s clearly has more advantages in terms of embedded requirements. This vision-assisted system capable of classifying chemical containers accurately in the scenarios of real-time chemical experiments will provide a good vision solution in the frontier fields of automated machine chemistry.

Improving the Data Quality of Untargeted Metabolomics through a Targeted Data-Dependent Acquisition Based on an Inclusion List of Differential and Preidentified Ions.

Metabolomics based on high-resolution mass spectrometry has become a powerful technique in biomedical research. The development of various analytical tools and online libraries has promoted the identification of biomarkers. However, how to make mass spectrometry collect more data information is an important but underestimated research topic. Herein, we combined full-scan and data-dependent acquisition (DDA) modes to develop a new targeted DDA based on the inclusion list of differential and preidentified ions (dpDDA). In this workflow, the MS1 datasets for statistical analysis and metabolite preidentification were first obtained using full-scan, and then, the MS/MS datasets for metabolite identification were obtained using targeted DDA of quality control samples based on the inclusion list. Compared with the current methods (DDA, data-independent acquisition, targeted DDA with time-staggered precursor ion list, and iterative exclusion DDA), dpDDA showed better stability, higher characteristic ion coverage, higher differential metabolites' MS/MS coverage, and higher quality MS/MS spectra. Moreover, the same trend was verified in the analysis of large-scale clinical samples. More surprisingly, dpDDA can distinguish patients with different severities of coronary heart disease (CHD) based on the Canadian Cardiovascular Society angina classification, which we cannot distinguish through conventional metabolomics data collection. Finally, dpDDA was employed to differentiate CHD from healthy control, and targeted metabolomics confirmed that dpDDA could identify a more complete metabolic pathway network. At the same time, four unreported potential CHD biomarkers were identified, and the area under the receiver operating characteristic curve was greater than 0.85. These results showed that dpDDA would expand the discovery of biomarkers based on metabolomics, more comprehensively explore the key metabolites and their association with diseases, and promote the development of precision medicine.

Boundary representation compatible feature recognition for manufacturing CAD models

This paper presents a Boundary Representation (BREP) compatible data representation for Graph Neural Network (GNN) based feature identification. This data representation follows BREP and STEP AP 203 standards and can transfer holistic manufacturing CAD information to the deep neural network, which assists to identify small local geometrical features and highly complex interactive geometric features. Inversely, this data representation can be easily converted back to a conventional CAD model due to its direct encoding approach. With this data representation, the GNN can reach 99.57% accuracy on 36 classes FeatureNet + dataset and realize 99.12% accuracy on Machining Process Identification (MPI) dataset with highly interactive features. This method can largely benefit Computer Assisted Process Planning and pave the way for future industrial automation.

Degramnet: effective audio analysis based on a fully learnable time–frequency representation

Current state-of-the-art audio analysis algorithms based on deep learning rely on hand-crafted Spectrogram-like audio representations, that are more compact than descriptors obtained from the raw waveform; the latter are, in turn, far from achieving good generalization capabilities when few data are available for the training. However, Spectrogram-like representations have two main limitations: (1) The parameters of the filters are defined a priori, regardless of the specific audio analysis task; (2) such representations do not perform any denoising operation on the audio signal, neither in the time domain nor in the frequency domain. To overcome these limitations, we propose a new general-purpose convolutional architecture for audio analysis tasks that we call DEGramNet, which is trained with audio samples described with a novel, compact and learnable time–frequency representation that we call DEGram. The proposed representation is fully trainable: Indeed, it is able to learn the frequencies of interest for the specific audio analysis task; in addition, it performs denoising through a custom time–frequency attention module, which amplifies the frequency and time components in which the sound is actually located. It implies that the proposed representation can be easily adapted to the specific problem at hands, for instance giving more importance to the voice frequencies when the network needs to be used for speaker recognition. DEGramNet achieved state-of-the-art performance on the VGGSound dataset (for Sound Event Classification) and comparable accuracy with a complex and special-purpose approach based on network architecture search over the VoxCeleb dataset (for Speaker Identification). Moreover, we demonstrate that DEGram allows to achieve high accuracy with lightweight neural networks that can be used in real-time on embedded systems, making the solution suitable for Cognitive Robotics applications.

Solar Photovoltaic System Performance Improvement Using a New Fault Identification Technique

Identifying faults in the photovoltaic (PV) arrays is very much essential in improving the PV system’s safety and reliability. Solar PV arrays operate with non-linear characteristics, installed with maximum power point trackers (MPPT’s), and blocking diodes cause mismatch levels. Line-to-line and line-to-ground faults are identified, and the faulted circuits are isolated by means of over current protection devices (OCPD) and ground fault protection devices (GFPD). In order to improve the accuracy of fault detection, artificial intelligence (AI)-based techniques like Fuzzy inference, wavelet, support vector machine, and k-nearest neighbors are used. The drawback of AI-based techniques are (1) requirement of large dataset for effective fault identification and also show incompatibility if there is low irradiation and (2) require a larger number of voltage and current sensors. An experimental setup of 160 W, 4 × 4 solar PV array having PV modules (SPB) is subjected to different fault conditions (CS), and the faults are identified using the minimum number of sensors. The faults that are not detected by the conventional methods are detected using this proposed method, and the power gain due to the fault identification is around 152% which is 97 W in the PV array.

COI barcoding provides reliable species identification and pinpoints cryptic diversity in Western Palearctic amphibians

Abstract Assembling DNA barcode reference libraries for various taxonomic groups allows researchers to use metabarcoding or environmental DNA approaches to gain a rapid understanding of diversity in given environments. However, our ability to use reference libraries depends on how accurately DNA barcodes are able to recover taxonomic boundaries and identify species, which is rarely considered. We constructed an extensive COI barcoding library for amphibians of the Western Palearctic and successfully recovered barcodes from 60 urodele and 73 anuran species (representing 94% and 98% of the nominal anuran and urodele species in the Western Palearctic, respectively), covering the intraspecific diversity of the majority of species in this region. We tested the effectiveness of our assembled DNA barcode dataset for species identification using barcoding gap, efficiency analyses, and two phylogenetic species delimitation methods. We obtained DNA barcodes for 1251 specimens (691 anurans and 560 urodeles) with a high success rate (92-96%) of species identification. The absence of a barcoding gap in a number of samples was linked to species misidentifications, which suggest incipient speciation or cryptic diversity, or previously described mitochondrial introgression events. The phylogenetic species delimitation methods resulted in substantial oversplitting of currently accepted taxonomy. This COI barcoding library provides an almost complete and reliable reference library for Western Palearctic amphibians. We highlight the importance of generating comprehensive and well curated reference libraries that include intra- and interspecific genetic variability and the need of detailed taxonomic revision when ambiguous or incorrect DNA barcodes exist.

A dataset of fine-grained fossils of the conodont genus Hindeodus for classification using convolutional neural networks

With the rise of artificial intelligence, the booming application of convolutional neural networks to the classification and identification of fossils has attracted more and more attention. According to our survey, it is found that the species classified by previous authors basically belong to different genera, families or higher biological taxonomic units. However, in fact, the identification of fossils between species within a genus is often the focus and challenge for the identification task, which means that the previously trained classifiers may not be suitable for actual fossil identification. On this basis, in this paper, we built a dataset covering 12 species of the conodont genus Hindeodus by means of literature collection, while providing an augmented dataset of the original data. Since the dataset is fine-grained, users can train it by using convolutional neural network combined with fine-grained image feature extraction technology. In view of the deficiencies of the dataset such as small amount of data and unbalanced classes, it is suggested that users use stratified K-fold cross-validation, transfer learning and weighted loss function in the training task to solve the above problems. The dataset is aimed to add a fine-grained fossil dataset to the field of intelligent identification of biological fossils, which can be used as an experimental dataset for intelligent identification of fine-grained (species-level) fossils by convolutional neural networks. The fine-grained primitive followed by this dataset can also be used as a reference for the establishment of other fossil datasets.

EA-ConvNeXt: An Approach to Script Identification in Natural Scenes Based on Edge Flow and Coordinate Attention

In multilingual scene text understanding, script identification is an important prerequisite step for text image recognition. Due to the complex background of text images in natural scenes, severe noise, and common symbols or similar layouts in different language families, the problem of script identification has not been solved. This paper proposes a new script identification method based on ConvNext improvement, namely EA-ConvNext. Firstly, the method of generating an edge flow map from the original image is proposed, which increases the number of scripts and reduces background noise. Then, based on the feature information extracted by the convolutional neural network ConvNeXt, a coordinate attention module is proposed to enhance the description of spatial position feature information in the vertical direction. The public dataset SIW-13 has been expanded, and the Uyghur script image dataset has been added, named SIW-14. The improved method achieved identification rates of 97.3%, 93.5%, and 92.4% on public script identification datasets CVSI-2015, MLe2e, and SIW-13, respectively, and 92.0% on the expanded dataset SIW-14, verifying the superiority of this method.