Categories Of Features Research Articles

Unveiling Lung Diseases in CT Scan Images With a Hybrid Bio‐Inspired Mutated Spider‐Monkey and Crow Search Algorithm

ABSTRACTBio‐inspired computer‐aided diagnosis (CAD) has garnered significant attention in recent years due to the inherent advantages of bio‐inspired evolutionary algorithms (EAs) in handling small datasets with elevated precision and reduced computational complexity. Traditional CAD models face limitations as they can only be developed post‐outbreak, relying on datasets that become available during such events such as the COVID‐19 pandemic. The scarcity of data for emerging diseases poses a substantial challenge to achieving elevated precision with conventional deep‐learning algorithms. Furthermore, even when datasets are available, employing deep learning for class‐based classification is arduous, necessitating model retraining, in this paper, we propose a novel hybrid algorithm that leverages the strengths of the crow search algorithm (CSA) and the spider monkey optimization (SMO) algorithm to create an optimised spider monkey crow search (OSM‐CS) algorithm. We developed a CAD tool that maps each input CT image to a high‐dimensional vector by extracting four categories of features: high contrast, polynomial decomposition, textural, and pixel statistics. The proposed OSM‐CS algorithm is employed as a feature selection method. Our experimental results demonstrate the effectiveness of the OSM‐CS algorithm, achieving an impressive accuracy of 98.2% when coupled with an AdaBoost classifier for multi‐class classification and 99.93% for binary classification. This performance surpasses that of state‐of‐the‐art (SOTA) deep learning models and recently published hybrid algorithms, underscoring the potential of the OSM‐CS algorithm as a powerful tool in the realm of CAD.

Cable fault diagnosis with generalization capability using incremental learning and deep convolutional neural network

Power cable faults may exhibit dynamic and differential changes attributable to variations in operating times and environmental conditions. However, the existing artificial intelligence (AI)-based fault diagnosis methods are lack of capability to handle this complexity of fault development. They can only diagnose given faulty types and will malfunction when dealing with newly-emerged faults, which accounts for the lack of generalization capability for AI-based methods. Given that, this paper proposes a cable fault diagnosis method with generalization capability utilizing incremental learning and deep convolutional neural network (DCNN). The idea of knowledge distillation is introduced to reserve the diagnosis information of the DCNN for original faults, and meanwhile, the DCNN is also modified by the classification loss of newly-emerged faulty samples, which equips the DCNN with generalization capability. The category features are obtained utilizing the refined DCNN, and sent to the constructed classifier for accurate incremental fault diagnosis. Experiments are conducted with field and simulated data. The obtained results validate the feasibility and effectiveness of the proposed method.

Preference learning based on adaptive graph neural network for multi-criteria decision support

Intelligent decision-making assists decision-makers (DMs) in making choices through data analysis, model prediction, and automated processes. Central to this field are two key concepts: multi-criteria decision making (MCDM) and preference learning (PL). While MCDM and PL both aim to develop decision models that rank alternatives based on observed or revealed preferences, they diverge in focus. MCDM concentrates on the DMs' perspectives, whereas PL emphasizes model-driven approaches. This divergence presents significant challenges in integrating these methodologies, particularly in ensuring the integrated method remains scalable and interpretable amidst the complexity of decision scenarios. To bridge this gap, our study introduces the use of graph structures to frame decision problems and proposes a novel PL method employing graph neural network (GNN) for multi-criteria decision support. This method is anchored in the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) technique and combines an adaptive GNN model with a weight determination model. The GNN model updates embeddings from the alternative's criteria and category features, utilizing an attention mechanism to adaptively assess their importance. Concurrently, the weight determination model contains a weight neural network module to set objective criterion weights and a game theory-based module for calculating combined criterion weights. The method not only inherits the interpretability and intuitive appeal of decision models but also leverages the computational efficiency and high accuracy of machine learning. In experiments conducted on benchmark datasets, our method exhibits significant performance improvements, especially in ranking-related evaluation metrics, outperforming the best baseline by 5.78%.

Text-to-Image Generation Using Stack Generative Adversarial Networks (GANs) and Stable Diffusion Models

Text-to-image synthesis is a challenging task in artificial intelligence that involves generating realistic images from textual descriptions. StackGAN, [1] a generative model with a two-stage architecture, has proven to be effective in improving the quality and fidelity of synthesized images. In this study, we implement and evaluate the performance of StackGAN using the CIFAR-10 dataset, demonstrating how a multi-stage generation approach can produce high-quality images from textual labels. The experimental results show that StackGAN successfully captures the underlying features of the dataset categories and provides reasonable image fidelity.

Predictive Modelling of Liver Disease Using Biochemical Markers and K-Nearest Neighbors Algorithm

The incidence of liver cirrhosis-related deaths is on the rise due to increased alcohol consumption, chronic hepatitis infections, and obesity-related liver conditions. Early detection is critical for improving patient outcomes; however, female patients often experience delayed diagnosis. This study aims to develop a predictive model for liver disease using biochemical markers and to investigate gender disparities in diagnostic accuracy. A dataset of 584 patient records from NorthEast Andhra Pradesh, India, was utilized, comprising ten variables per patient, including age, gender, total bilirubin, direct bilirubin, alkaline phosphatase, SGPT, SGOT, total proteins, albumin, and the albumin/globulin ratio. The data were pre-processed by encoding categorical variables and scaling numerical features. The K-Nearest Neighbors (K-NN) algorithm was employed for classification, and performance was evaluated using cross-validation. The model demonstrated variable accuracy across different folds, with accuracy ranging from 57.76% to 73.28%, precision from 58.14% to 70.56%, recall from 57.76% to 73.28%, and F1-score from 57.95% to 70.45%. These results indicate the potential of biochemical markers in predicting liver disease and highlight significant gender disparities in diagnostic accuracy. The study's contributions include the development of a practical predictive tool and the identification of gender-specific diagnostic challenges. Future research should focus on larger, more diverse datasets and explore additional machine learning algorithms to enhance predictive accuracy and address gender disparities in liver disease diagnosis.

Sensory Design in Healthcare Welcome Spaces for Healthy Ageing.

Primary Care Services in Italy are evolving to promote the spread of social-health care facilities (e.g., CdC - Casa della Comunità/House of the Community). Many gaps have been detected in the field of designing for CdC, especially regarding designing welcome and waiting spaces. Sensory Design was identified as a suitable approach to design comfortable and customisable environments for the different CdC's users. Sensory environments aim at favouring mental health, rehabilitation and comfort in healthcare facilities, reducing stress during the waiting time and before or during the medical intervention as a positive distraction. Scientific literature is still not expanded in this field. The contribution aims at studying the application of Sensory Design in an innovative context such as the CdC environment to support healthy and active ageing. This design approach favours the stimulation of primary senses and self-regulated emotions to generate positive feelings, reduce stress, promote relaxation, physical activity and recovery. According to a Theoretical Framework, best practices have been analysed to identify spatial and environmental sensory characters to be applied in designing CdC's welcome and waiting spaces. The authors have identified four main categories of sensory features for CdC: i) innovative Spatial Models: designing the building according to a multisensory approach to stimulate and welcome people in an inclusive way; diffusing sensory equipment in public spaces (e.g., corridors, waiting spaces, etc.); designing sensory atrium or waiting spaces; providing snoezelen rooms as dedicated and specific rooms for sensory recovery; making environments adaptable trough portable sensory equipment; ii) Relaxing features to promote comfort and psychological support; iii) Physical activity promotion, according to the idea of healthy and active ageing; iv) the integration of sensory solutions by the use of digital technology.

Domain-Invariant Few-Shot Contrastive Learning for Hyperspectral Image Classification

In Hyperspectral Image (HSI) classification, acquiring large quantities of high-quality labeled samples is typically costly and impractical. Traditional deep learning methods are limited in such scenarios due to their dependence on sample quantities. To address this challenge, researchers have turned to Few-Shot Learning (FSL). Although existing FSL methods improve classification performance by enhancing domain invariance through domain adaptation, they often overlook the critical issue of high inter-class similarity and large intra-class variability. Moreover, during domain alignment, features of different categories within the same domain may become confused. To address these issues, this paper proposes a novel Domain-Invariant Few-Shot Contrastive Learning (DIFSCL) method, which combines domain adaptation and contrastive learning strategies to not only learn domain-invariant features but also significantly enhance inter-class discriminability. Based on this, we further design a multi-scale adaptive attention mechanism for a hyperspectral feature extraction network to more effectively extract and optimize generalized features within the DIFSCL framework, significantly improving intra-class consistency and inter-class discriminative capability of features. Experimental results on three widely used HSI datasets demonstrate that our method significantly outperforms existing techniques in few-shot classification tasks.

A semantic consistent object detection model for domain adaptation based on mixed-class distribution metrics

Unsupervised domain adaptation is crucial for mitigating the performance degradation caused by domain bias in object detection tasks. In previous studies, the focus has been on pixel-level and instance-level shift alignment to minimize domain discrepancy. However, it is important to note that this method may inadvertently align single-class instance features with mixed-class instance features that belong to multiple categories within the same image during image-level domain adaptation. This challenge arises because each image in object detection tasks contains objects of multiple categories. To achieve the same category feature alignment between single-class and mixed-class, our method considers features with different mixed categories as a new class and proposes a mixed-classes H-divergence to reduce domain bias for object detection. To enhance both single-class and mixed-class semantic information, and to achieve semantic separation for the mixed-classes in H-divergence, we employ Semantic Prediction Models (SPM) and Semantic Bridging Components (SBC). Furthermore, we reweigh the loss of the pixel domain discriminator based on the SPM results to reduce sample imbalance. Our extensive experiments on widely used datasets illustrate how our method can robustly improve object detection in domain bias settings.

Testing relationships between multiple regional features and biogeographic processes of speciation, extinction, and dispersal.

The spatial and environmental features of regions where clades are evolving are expected to impact biogeographic processes such as speciation, extinction, and dispersal. Any number of regional features (such as elevation, distance, area, etc.) may be directly or indirectly related to these processes. For example, it may be that distances or differences in elevation or both may limit dispersal rates. However, it is difficult to disentangle which features are most strongly related to rates of different processes. Here, we present an extensible Multi-feature Feature-Informed GeoSSE (MultiFIG) model that allows for the simultaneous investigation of any number of regional features. MultiFIG provides a conceptual framework for incorporating large numbers of features of different types, including categorical, quantitative, within-region, and between-region features, along with a mathematical framework for translating those features into biogeographic rates for statistical hypothesis testing. Using traditional Bayesian parameter estimation and reversible-jump Markov chain Monte Carlo, MultiFIG allows for the exploration of models with different numbers and combinations of feature-effect parameters, and generates estimates for the strengths of relationships between each regional feature and core process. We validate this model with a simulation study covering a range of scenarios with different numbers of regions, tree sizes, and feature values. We also demonstrate the application of MultiFIG with an empirical case study of the South American lizard genus Liolaemus, investigating sixteen regional features related to area, distance, and elevation. Our results show two important feature-process relationships: a negative distance/dispersal relationship, and a negative area/extinction relationship. Interestingly, although speciation rates were found to be higher in Andean versus non-Andean regions, the model did not assign significance to Andean- or elevation-related parameters. These results highlight the need to consider multiple regional features in biogeographic hypothesis testing.

H-index and research evaluation: A suggested set of components for developing a comprehensive author-level index

The H-index has been investigated in various studies; this index has many strengths that have made it popular. However, it also has weaknesses, due to which other indicators have been developed. This study aims to identify the strengths and weaknesses of the H-index and provide the minimum set of necessary components for developing a comprehensive author-level index. In this systematic literature review, Scopus, PubMed, Web of Science, Emerald, and ProQuest databases were searched to identify relevant studies. From the number of 14,253 retrieved studies, after two stages of screening, 81 studies were selected according to the eligibility criteria for data extraction. The findings of the study led to the identification of 15 strengths in the three categories of Quality Features, Simplicity, and Suitability, and 13 weaknesses in the six categories of Publications, Citations, Academic Age, Author Credit Allocation, Variety of Fields, and mathematical calculation for H-index. Finally, 28 components were identified as the minimum set of necessary components to develop a comprehensive author-level index to help evaluate researchers more realistically and fairly. The minimum components that need to be considered in developing a comprehensive author-level index can be proposed as follows: Quality Features, Simplicity, Suitability, Publications, Citations, Academic Age, Author Credit Allocation, Variety of Fields, and mathematical calculation.

Automatic TNM staging of colorectal cancer radiology reports using pre-trained language models

Background and Objective:Colorectal cancer is one of the major causes of cancer death worldwide. Essential for prognosis and treatment planning, TNM staging offers critical insights into the advancement of colorectal cancer. However, manual TNM staging from colon magnetic resonance imaging (MRI) is a laborious and error prone process. This study introduces an automated text classification system for TNM staging of colon MRI images in Spanish. Methods:A dataset of 1319 Spanish colon MRI reports was collected and manually labeled with TNM staging. In order to automate the task of TNM staging, a multimodal system was proposed. The system is based on RoBERTa language model pre-trained on a combination of biomedical and clinical Spanish language corpora and uses Natural Language Processing (NLP) techniques to extract relevant categorical and numerical features from MRI reports. Results:The performance of the system was evaluated using different metrics and the results obtained are very promising: the best performance among the proposed systems reached 0.7464, 0.8792 and 0.6776 of macro F1-score for T, N and M respectively. Conclusions:This study demonstrates the feasibility of using a language model for automatic TNM staging based on Spanish clinical reports of colorectal cancer patients. The proposed system can be a useful tool to improve the efficiency and accuracy of colorectal cancer diagnosis.

A generalized health index: automated thoracic CT-derived biomarkers predict life expectancy.

To identify image biomarkers associated with overall life expectancy from low-dose computed tomography and integrate them as an index for assessing an individual's health. Two categories of CT image features, body composition tissues and cardiopulmonary vasculature characteristics, were quantified from LDCT scans in the Pittsburgh Lung Screening Study cohort(n = 3,635). Cox proportional-hazards models identified significant image features which were integrated with subject demographics to predict the subject's overall hazard. Subjects were stratified using composite model predictions and feature-specific risk stratification thresholds. The model's performance was validated extensively, including 5-fold cross-validation on PLuSS baseline, PLuSS follow-up examinations, and the National Lung Screening Trial (NLST). The composite model had significantly improved prognostic ability compared to the baseline model (p < 0.01) with AUCs of 0.774 (95% CI: 0.757-0.792) on PLuSS, 0.723 (95% CI: 0.703-0.744) on PLuSS follow-up, and 0.681 (95% CI: 0.651-0.710) on the NLST cohort. The identified high-risk stratum were several times more likely to die, with mortality rates of 79.34% on PLuSS, 76.47% on PLuSS follow-up, and 46.74% on NLST. Two cardiopulmonary structures (intrapulmonary artery vein ratio, intrapulmonary vein density) and two body composition tissues (SM density, bone density) identified high-risk patients. Body composition and pulmonary vasculatures are predictive of an individual's health risk; their integrations with subject demographics facilitate the assessment of an individual's overall health status or susceptibility to disease. CT-computed body composition and vasculature biomarkers provide improved prognostic value. The integration of CT biomarkers and patient demographic information improves subject risk stratification.

High-Order Multimodal Interaction Network for Efficient Prediction of Drug-Drug Interaction

The pharmaceutical industry places a lot of focus on preventing drug-drug interactions (DDIs). Medication interaction detection has been the primary emphasis of machine learning-based DDI prediction methods. Since studies have shown that DDIs can cause different future occurrences, it is more beneficial to predict DDI connected events while examining the mechanism behind combination medicine consumption or adverse reactions. The areas of medication development and illness diagnostics are seeing increased usage of a developing approach that predicts DDIs-associated occurrences. We examine potential interactions between the two medications as well as the kinds of interactions that can occur in this research. Additionally, we provide a strategy that relies on learning and use High-Order Multimodal Interaction Network (HOMIN) to forecast DDIs by learning feature representations. The purpose of this work is to provide HOMIN-DDI, a new method for drug-drug interaction predictions that we have developed utilizing a HOMIN architecture. First, we develop feature vectors from medication categories, targets, paths, and digestive enzymes, and we use the Jaccard similarity to evaluate how similar drugs appear. After that, we build a new HOMIN using the feature representation to predict DDIs. The results of the experiments show that the drug categories feature type, when used to the HOMIN -DDI approach, is effective. Furthermore, compared to employing only one feature, combining numerous features yields more useful and informative results. Compared to other algorithms, HOMIN-DDI performs better when it comes to predicting DDIs.

Deep transfer learning for microseismic waveforms recognition across geological conditions in TBM tunnels

In deeply buried tunneling projects, geological conditions are often complex and varied. Microseismic monitoring systems are extensively deployed to enhance construction safety. However, when the current geological conditions differ from those present during the signal collection for model training, recognition accuracy tends to decline significantly. Therefore, improving the applicability and stability of microseismic waveform recognition models across varying geological conditions has emerged as a critical challenge. To address this issue, we first analyze the impact of lithological changes and the development of structural planes on the features of microseismic waveforms. Subsequently, we propose a category-domain-aligned transfer learning method that enables the transfer of recognition capabilities across geological conditions by facilitating similar feature extraction and the recognition of cross-geological fracture waveforms. In this model, feature separation modeling enhances the extraction of category features of waveforms under different geological conditions. A deep transfer learning mechanism distinguishes between unique and common features, allowing for the capture of essential features necessary for model parameter updates. Through comparative experiments and feature distribution alignment and visualization, we demonstrate that the accuracy of microseismic waveform recognition across geological conditions achieves 90 %. Additionally, the performance of our method is validated using microseismic signals collected from different sections of the construction site.

Using topological data analysis and machine learning to predict customer churn

Customer churn is a common problem faced by many industries, including telecommunication industries. This has resulted in the development of advanced techniques for the prediction and prevention of customer churn. The availability of stored customer data in the form of big data, together with the use of advanced and tuned machine learning (ML) algorithms, have paved the way for the realisation and extraction of useful features associated with customer behaviour and consequently the prediction of customer churn. An effective way to further improve churn prediction capability of different ML algorithms is through the employment of topological data analysis (TDA). TDA is a framework that applies topological methods to uncover the underlying hidden structural features in complex, high-dimensional data. Here, a TDA summary of 0- and 1-dimensional holes of the data, called barcode statistics, was extracted and used as an additional feature to the preprocessed customer data. To address issues such as the effective preprocessing and analysis of large customer datasets and the effective tuning of ML hyperparameters, we implement an advanced data preprocessing technique that consists of different stages such as handling of missing data, feature engineering, encoding of categorical features using the hashing encoding method, and feature selection. Without including barcode statistics in the model, the XGBoost algorithm with tuned hyperparameters achieved the best results, with accuracy of 92.71%, precision of 85.95%, recall of 92.71%, and F-measure of 89.20%. Including barcode statistics as an additional feature, the XGBoost algorithm with tuned hyperparameters achieved the best and much improved results, with accuracy of 98.50%, precision of 98.50%, recall of 98.50%, and F-measure of 98.50%. The use of TDA barcode statistics significantly improved the churn prediction capability of the ML algorithms. In addition, hyperparameter tuning is not needed when an effective data preprocessing technique is used, or when barcode statistics is used. The best accuracy of 98.5% from this work was in line with the best accuracy of 98.7% from a related work, but interestingly, the best precision of 98.5% from this work was superior to the 94.3% precision from the same related work with higher accuracy.

CFFANet: category feature fusion and attention mechanism network for retinal vessel segmentation

CFFANet: category feature fusion and attention mechanism network for retinal vessel segmentation

Clinical and epidemiological characteristics of patients with pheochromocytoma and retroperitoneal paraganglioma

Introduction. Pheochromocytoma/paraganglioma (PPGL) is a rare hormonally active tumor of chromaffin tissue. Delayed diagnosis of this disease can lead to severe, even fatal, complications. Surgical treatment of PPGL without the necessary preoperative preparation can cause intraoperative uncontrolled hemodynamic impairment. In addition, since 2017, PPGL has been recognized as a malignant tumor due to the presence of metastatic potential. All these factors determine the need for early diagnosis of PPGL, which is based, first of all, on knowledge of the clinical picture by specialists in various fields.Aim. To check and confirm the epidemiological characteristics of PPGL as well as the most common clinical symptoms in the whole group of PPGL and in subgroups of clinical forms; to form a risk group for screening for PPGL.Materials and methods. There were enrolled 353 patients in the study. These patients were operated on at the St. Petersburg State University Hospital in the period from 2010 to 2022 inclusive. All the patients were verified PPGL diagnosis histologically. Clinical symptoms were assessed: a) retrospectively, by analyzing complaints and anamnesis from the initial examination in medical histories upon admission to the hospital for surgical treatment b) both in the entire group as a whole and in subgroups of clinical forms (paroxysmal, constant, mixed, asymptomatic). Statistical analysis was performed using the Python 3.11 programming language (Python Software Foundation, USA). descriptive statistics of quantitative characteristics are presented in the form of medians, first and third quartiles in me format [q1; q3], categorical features – in the form of absolute and relative frequencies in the format n (%). the frequency histogram was constructed using the Matplotlib 3.8.2 library.Results. PPGL is most common in women over 40 years of age. The most common symptoms of PPGL are the following: a feeling of rapid heartbeat (55.95%), headache (41.07%), arterial hypertension (AH) in the form of hypertensive crises (55.65%), constant increase in blood pressure (38.39%), episodes of moderate increases in blood pressure (24.11%), sweating (30.65%), general weakness and fatigue (26.19%), hot flashes or chills (24.11%). These symptoms are formed into the criteria for screening for PPGL. In the asymptomatic clinical form – lower back pain is more common (14%); with paroxysmal – panic attacks (10%); with mixed – thirst and dry mouth (10%), and with constant form – the actual constant nature of hypertension (100%).Conclusion. In a large sample of patients with a rare pathology, the most characteristic symptom complex was identified, characteristic of both PPGL and individual clinical forms. A risk group has been formed for screening for PPGL.

Integrating VGG 19 U‐Net for Breast Thermogram Segmentation and Hybrid Enhancement With Optimized Classifier Selection: A Novel Approach to Breast Cancer Diagnosis

ABSTRACTEarly diagnosis of breast cancer is essential for improving patient survival rates and reducing treatment costs. Despite breast thermogram images having high quality, doctors in developing countries often struggle with early diagnosis due to difficulties in interpreting subtle details. Implementing a Computer‐Aided Diagnosis (CAD) system can assist doctors in accurately analyzing these details. This article presents an innovative approach to breast cancer diagnosis using thermal images. The proposed method enhances the quality and clarity of relevant features while preserving sharp and curved edges through U‐Net‐based segmentation for automatic selection of the ROI, advanced hybrid image enhancement techniques, and a machine learning classifier. Subjective analysis compares the processed images with five conventional enhancement techniques, demonstrating the efficiency of the proposed method. The quantitative analysis further validates the effectiveness of the proposed method against five conventional methods using four quality measures. The proposed method achieves superior performance with PSNR of 15.27 for normal and 14.31 for malignant images, AMBE of 6.594 for normal and 7.46 for malignant images, SSIM of 0.829 for normal and 0.80 for malignant images, and DSSIM of 0.084 for normal and 0.14 for malignant images. The classification phase evaluates four classifiers using 13 features from three categories. The Random Forest (RF) classifier with Discrete Wavelet Transform (DWT) based features initially outperformed other classifier features but had limited performance, with accuracy, sensitivity and specificity of 81.8%, 88.8%, and 91%, respectively. To improve this, three categories of features were normalized and converted into two principal components using Principal Component Analysis (PCA) to train the RF classifier, which then showed superior performance with 97.7% accuracy, 96.5% sensitivity, and 98.2% specificity. The dataset utilized in this article is obtained from the Indira Gandhi Centre for Atomic Research (IGCAR), Kalpakkam, India. The entire proposed model is implemented in a Jupyter notebook.

Dual Attention Relation Network With Fine-Tuning for Few-Shot EEG Motor Imagery Classification.

Recently, motor imagery (MI) electroencephalography (EEG) classification techniques using deep learning have shown improved performance over conventional techniques. However, improving the classification accuracy on unseen subjects is still challenging due to intersubject variability, scarcity of labeled unseen subject data, and low signal-to-noise ratio (SNR). In this context, we propose a novel two-way few-shot network able to efficiently learn how to learn representative features of unseen subject categories and classify them with limited MI EEG data. The pipeline includes an embedding module that learns feature representations from a set of signals, a temporal-attention module to emphasize important temporal features, an aggregation-attention module for key support signal discovery, and a relation module for final classification based on relation scores between a support set and a query signal. In addition to the unified learning of feature similarity and a few-shot classifier, our method can emphasize informative features in support data relevant to the query, which generalizes better on unseen subjects. Furthermore, we propose to fine-tune the model before testing by arbitrarily sampling a query signal from the provided support set to adapt to the distribution of the unseen subject. We evaluate our proposed method with three different embedding modules on cross-subject and cross-dataset classification tasks using brain-computer interface (BCI) competition IV 2a, 2b, and GIST datasets. Extensive experiments show that our model significantly improves over the baselines and outperforms existing few-shot approaches.

Reconstructing urban vegetation evolution in China using multimodal deep learning and 30-years Landsat archive

To retrospect the impact of urbanization on vegetation, consistent and reliable maps that provide urban vegetation estimates are required. In this study, we make the first attempt to employ deep learning for urban vegetation mapping of 2120 cities in China using a 30-years Landsat archive (UV-30). We present a multimodal deep learning (MDL) model to combine features from 5 categories, namely reflective bands, remote sensing indices, texture variables, topographical variables, and time labels. We normalized elevation data locally to remove the inter-city differences, while retaining topographic details. We used time labels to account for spectral variability over time and Landsat sensors. The performance of the MDL models was compared with a vanilla deep neural network, random forest, and support vector regression. Results indicated that the fusion of different categories of features improved the prediction accuracy. The MDL performed best in most validation cities, and our results were superior to the fractional maps aggregated from high-resolution datasets in homogeneous urban landscapes. We further analyzed urban vegetation dynamics across the urban core-new town-outskirt gradients of different geographical regions and city sizes. We found a gradually diminishing de-greening trend in urban China, especially in urban cores and new towns. Nevertheless, we observed a more pronounced degradation of urban vegetation in the south regions of China compared to the north, and smaller cities show no evident urban vegetation increase. The suggested framework is valuable for urban vegetation assessments and facilitates a thorough understanding and management of urban ecosystems.