GeoAI in terrain analysis: Enabling multi-source deep learning and data fusion for natural feature detection

Rice diseases are major threats to rice yield and quality. Rapid and accurate detection of rice diseases is of great importance for precise disease prevention and treatment. Various spectroscopic techniques have been used to detect plant diseases. To rapidly and accurately detect three different rice diseases [leaf blight (Xanthomonas oryzae pv. Oryzae), rice blast (Pyricularia oryzae), and rice sheath blight (Rhizoctonia solani)], three spectroscopic techniques were applied, including visible/near-infrared hyperspectral imaging (HSI) spectra, mid-infrared spectroscopy (MIR), and laser-induced breakdown spectroscopy (LIBS). Three different levels of data fusion (raw data fusion, feature fusion, and decision fusion) fusing three different types of spectral features were adopted to categorize the diseases of rice. Principal component analysis (PCA) and autoencoder (AE) were used to extract features. Identification models based on each technique and different fusion levels were built using support vector machine (SVM), logistic regression (LR), and convolution neural network (CNN) models. Models based on HSI performed better than those based on MIR and LIBS, with the accuracy over 93% for the test set based on PCA features of HSI spectra. The performance of rice disease identification varied with different levels of fusion. The results showed that feature fusion and decision fusion could enhance identification performance. The overall results illustrated that the three techniques could be used to identify rice diseases, and data fusion strategies have great potential to be used for rice disease detection.

Recently, tremendous interest has been devoted to develop data fusion strategies for energy efficiency in buildings, where various kinds of information can be processed. However, applying the appropriate data fusion strategy to design an efficient energy efficiency system is not straightforward; it requires a priori knowledge of existing fusion strategies, their applications and their properties. To this regard, seeking to provide the energy research community with a better understanding of data fusion strategies in building energy saving systems, their principles, advantages, and potential applications, this paper proposes an extensive survey of existing data fusion mechanisms deployed to reduce excessive consumption and promote sustainability. We investigate their conceptualizations, advantages, challenges and drawbacks, as well as performing a taxonomy of existing data fusion strategies and other contributing factors. Following, a comprehensive comparison of the state-of-the-art data fusion based energy efficiency frameworks is conducted using various parameters, including data fusion level, data fusion techniques, behavioral change influencer, behavioral change incentive, recorded data, platform architecture, IoT technology and application scenario. Moreover, a novel method for electrical appliance identification is proposed based on the fusion of 2D local texture descriptors, where 1D power signals are transformed into 2D space and treated as images. The empirical evaluation, conducted on three real datasets, shows promising performance, in which up to 99.68% accuracy and 99.52% F1 score have been attained. In addition, various open research challenges and future orientations to improve data fusion based energy efficiency ecosystems are explored.

The complexity and diversity of equipments presented in the development of modern industrial technology brings the wide application of deep learning in fault diagnosis. However, the multi-source heterogeneous data collected from different sensors applied in industrial production cannot be used by traditional neural networks directly, which registers a difficulty to fault diagnosis based on deep learning. In addition, with only one of the data used, traditional deep learning fault diagnosis methods ignores the correlation between heterogeneous data, which may result in the loss of useful information and the accuracy of fault diagnosis. To solve the above problems, a fault diagnosis framework based on deep learning for multi-source heterogeneous data fusion is proposed. In the feature learning stage, stack self-encoder (SAE) and convolution neural network (CNN) are used to extract deep features from one-dimensional vibration data and two-dimensional image data respectively, and then feature fusion fault diagnosis is carried out. It solves the problem that traditional neural network can't make full use of multi-source heterogeneous data, and improves the accuracy of fault diagnosis. What's more, the bearing data of Case Western Reserve University are used to verify the validity of the diagnosis method proposed in this paper.

1H NMR and UV-visible data fusion for determining Sudan dyes in culinary spices

HighlightsE-nose and computer vision combined with data fusion strategies were applied to trace tea origins.Pearson correlation analysis, IG, and F-scores were applied to modify the fusion strategies.The classification performances of different fusion strategies were compared.The strategies of IG_SVM_FL and IG_SVM_DS achieved the best results.Abstract. The traceability of tea origins is of great significance. In this study, an electronic nose (E-nose) and computer vision system (CVS) were jointly applied to acquire aroma and image signals of tea samples, aiming at identifying Longjing teas from different geographic origins including Jinyun (120° 7' E, 28° 65' N), Xihu (120° 13' E, 30° 27' N), Xinchang (120° 9' E, 29° 50' N), and Qian Daohu (119° 3' E, 29° 60' N). Data fusion was used to integrate the E-nose and CVS signals for comprehensively characterizing the tea samples. Four traditional fusion strategies including k-nearest neighbors (KNN) and support vector machine (SVM) based feature-level fusion (KNN_FL and SVM_FL) and Dempster-Shafer (D-S) evidence theory based decision-level strategies (KNN_DS and SVM_DS) were applied for classification modeling. Pearson analysis, information gain (IG), and F-scores were employed to modify the traditional fusion strategies to reduce inconsistent and redundant information in the fusion process. The results indicated that the original fusion strategies had no superiority over independent E-nose and CVS decision-making. With the feature selection methods, the modified fusion strategies generally exhibited better performance than the independent decision-making and original fusion strategies. Moreover, the IG-based fusion strategies, encompassing IG_SVM_FL and IG_SVM_DS, achieved the highest classification accuracy of 100%. Keywords: Computer vision, Electronic nose, Feature selection, Fusion strategies, Tea origins.

This systematic literature review examines recent developments in stock market prediction using heterogeneous data sources that combine technical indicators, fundamental attributes, and sentiment-driven signals. Despite the growing adoption of machine learning in financial forecasting, existing research remains fragmented across data modalities, fusion strategies, and evaluation protocols, limiting comparability and practical applicability. Studies published between 2018 and 2024 were retrieved from five major scholarly databases and screened based on predefined eligibility criteria, resulting in 44 peer-reviewed articles included in the final analysis. The review synthesizes the quantitative and qualitative data modalities employed, the machine learning and deep learning methodologies adopted, the evaluation metrics used to assess predictive performance, and the principal challenges associated with multi-source stock market prediction. Findings reveal a clear shift toward deep learning architectures, hybrid fusion techniques, and the integration of external information such as news, corporate disclosures, and social media sentiment. Despite this progress, the literature exhibits inconsistent evaluation practices, limited attention to temporal data leakage, and insufficient coverage of non-English and emerging markets. This review consolidates current knowledge, presents a structured taxonomy of heterogeneous data sources and fusion strategies, and identifies open research challenges to guide future work in multimodal stock market prediction.

The review paper discusses the development and assessment of hybrid machine learning frameworks for early disease diagnosis using multi-source clinical data. It highlights the importance of early disease forecasting in healthcare, as it allows for more efficient illness management, mitigating symptom severity, and decelerating disease development. Traditional disease prediction methods often face challenges, such as reliance on isolated data sources and the existence of imbalanced datasets. Hybrid machine learning models offer a robust approach to address these shortcomings by integrating the advantages of multiple machine learning methods to enhance predictive accuracy and resilience. Case studies have demonstrated the effectiveness of hybrid machine learning models in predicting coronary heart issues, Alzheimer's brain related issues, diabetes, and lung cancer. This review study discusses various studies on the use of deep learning models in medical image analysis, heart disease prediction, and brain tumor detection. The study discusses various methods, such as hybrid machine learning, ensemble machine learning, multi-source transfer learning, convolutional neural networks, and cloud-enabled access control models. It also discusses the role of advanced learning models such as AI and deep learning in early detection of chronic diseases and the spread infection-based diseases. The study also discusses the challenges and opportunities in using machine learning in cardiovascular disease prediction, with some focusing on the use of jellyfish optimization algorithm and others on the use of boosting techniques. The review concludes by focus attention on the importance of understanding the role of machine learning in healthcare and the potential for future advancements in this field.

Focusing on the problem of relational data fusion in the environment with “information isolated island”, this paper presents a multi-sources relational data fusion (MSF) framework. The framework consists of three components: schema matching, entity alignment, and entity fusion. Based on the Hungarian algorithm, we propose an alignment discovery mechanism for the attributes alignment among multi-sources relational data. By extracting the multi-dimensional features of attribute values, we efficiently realized schema matching of multi-sources relational data. To link the tuple pairs from multi-source data, we introduced the diversity sampling strategy and the entity feature extraction approach. These can effectively improve the performance of entity alignment. Finally, linked entities are fused to provide a unified view of data analysis. To verify the usefulness and efficiency of the proposed methods, we implemented a fusion system called Data Puzzle, which is verified with the real public multi-field data. Experimental results demonstrate that the proposed methods can fuse multi-source relational data efficiently with high recall and precision.

Panax notoginseng (Burk.) F.H. Chen (P. notoginseng), a precious medicinal herb, provides both considerable values. There is confusion in the market regarding the geographical origins of P. notoginseng. It is of paramount importance to authenticate geographical origins of P. notoginseng, which will facilitate the rational utilization of resources and sustainable development. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were employed to investigate the influence of the tissue and data fusion strategy for the authentication of P. notoginseng origins. The PLS-DA model employing the main root spectra demonstrated great potential for authenticating P. notoginseng origin. The PLS-DA training and test sets based on feature-level fusion (VIP) spectral data were 100% accurate, indicating excellent potential in authenticating P. notoginseng. The data fusion strategy improved the authentication. The article presents a reliable and efficacious novel methodology for authenticating P. notoginseng origins and a scientific foundation for its quality control.

As a list of remotely sensed data sources is available, how to efficiently exploit useful information from multisource data for better Earth observation becomes an interesting but challenging problem. In this paper, the classification fusion of hyperspectral imagery (HSI) and data from other multiple sensors, such as light detection and ranging (LiDAR) data, is investigated with the state-of-the-art deep learning, named the two-branch convolution neural network (CNN). More specific, a two-tunnel CNN framework is first developed to extract spectral-spatial features from HSI; besides, the CNN with cascade block is designed for feature extraction from LiDAR or high-resolution visual image. In the feature fusion stage, the spatial and spectral features of HSI are first integrated in a dual-tunnel branch, and then combined with other data features extracted from a cascade network. Experimental results based on several multisource data demonstrate the proposed two-branch CNN that can achieve more excellent classification performance than some existing methods.

With the acceleration of intelligent transformation of energy system, the monitoring of equipment operation status and optimization of production process in thermal power plants face the challenge of multi-source heterogeneous data integration. In view of the heterogeneous characteristics of physical sensor data, including temperature, vibration and pressure that generated by boilers, steam turbines and other key equipment and real-time working condition data of SCADA system, this paper proposes a multi-source heterogeneous data fusion and analysis platform for thermal power plants based on edge computing and deep learning. By constructing a multi-level fusion architecture, the platform adopts dynamic weight allocation strategy and 5D digital twin model to realize the collaborative analysis of physical sensor data, simulation calculation results and expert knowledge. The data fusion module combines Kalman filter, wavelet transform and Bayesian estimation method to solve the problem of data time series alignment and dimension difference. Simulation results show that the data fusion accuracy can be improved to more than 98%, and the calculation delay can be controlled within 500 ms. The data analysis module integrates Dymola simulation model and AERMOD pollutant diffusion model, supports the cascade analysis of boiler combustion efficiency prediction and flue gas emission monitoring, system response time is less than 2 seconds, and data consistency verification accuracy reaches 99.5%.

Parkinson’s disease (PD) is a serious neurodegenerative disease. Most of the current treatment can only alleviate symptoms, but not stop the progress of the disease. Therefore, it is crucial to find medicines to completely cure PD. Finding new indications of existing drugs through drug repositioning can not only reduce risk and cost, but also improve research and development efficiently. A drug repurposing method was proposed to identify potential Parkinson’s disease-related drugs based on multi-source data integration and convolutional neural network. Multi-source data were used to construct similarity networks, and topology information were utilized to characterize drugs and PD-associated proteins. Then, diffusion component analysis method was employed to reduce the feature dimension. Finally, a convolutional neural network model was constructed to identify potential associations between existing drugs and LProts (PD-associated proteins). Based on 10-fold cross-validation, the developed method achieved an accuracy of 91.57%, specificity of 87.24%, sensitivity of 95.27%, Matthews correlation coefficient of 0.8304, area under the receiver operating characteristic curve of 0.9731 and area under the precision–recall curve of 0.9727, respectively. Compared with the state-of-the-art approaches, the current method demonstrates superiority in some aspects, such as sensitivity, accuracy, robustness, etc. In addition, some of the predicted potential PD therapeutics through molecular docking further proved that they can exert their efficacy by acting on the known targets of PD, and may be potential PD therapeutic drugs for further experimental research. It is anticipated that the current method may be considered as a powerful tool for drug repurposing and pathological mechanism studies.

The advancement of remote sensing technology aids geologists in obtaining lithological maps more quickly, comprehensively, and accurately. However, key challenges in lithological mapping include the limited spectral information from individual sensors and the difficulties in visually interpreting lithological samples. In this study, we integrated 241 scenes of optical data and 106 scenes of radar data on the Google Earth Engine (GEE) platform, proposing a rapid lithological identification framework that combines an automatic lithological sample data generation strategy with multi-source data. Using various machine learning algorithms, we evaluated the classification capabilities of heterogeneous predictive factors, feature optimization algorithms, and object-based algorithms. Results indicate that: (1) Combining optical and radar data improves prediction accuracy, with terrain data further enhancing mapping capabilities; (2) Terrain factors contribute most to classification, but SWIR and TIR bands of optical data are critical for lithological identification; (3) The feature optimization algorithm reduces feature redundancy and efficiency issues from multi-source data, achieving 96.51% accuracy with the optimal feature model, an improvement of 0.1%−2.02% over original features; (4) Object-based algorithms show significant potential in mapping areas with large rock outcrops. This study offers new insights for medium- to large-scale lithological maps and provides essential data support for geological work.

Tool wear monitoring is the key part of intelligent maintenance and has been attracting considerable interest. However, traditional data-driven methods assume that the collected data following identical distribution and the training data is sufficient, which is impractical in practice. This paper proposed a novel framework to realize tool wear monitoring across scenarios based on deep transfer learning. The collected multi-source heterogeneous signal is fused based on the feature self-extraction and selection capabilities of deep learning. Particularly, the feature extraction capability of convolutional neural networks (CNN) for structured data is improved by integrating traditional feature extraction methods. Furthermore, the transfer learning technique is introduced to migrate the pre-trained model from the source domain to the target domain, and thus achieving the tool wear monitoring across scenarios. The proposed framework was applied to the continuous cutting process of tools and the excellent experimental results demonstrated the effectiveness of the deep transfer learning network for tool wear monitoring with a small number of labeled data, which demonstrates the practicality of the proposed framework.

To improve the effect of college English teaching, combines multi-source data (MSD) fusion technology to build a college English blended teaching system. Moreover, through the research on the functions of smarter classrooms and the analysis and exploration of blended teaching, this paper designs a college English blended teaching system based on MSD fusion. In addition, this paper analyzes the bandwidth-tunable optical routing module of the interference-coupled series double-ring involved in multi-data fusion. Finally, this paper studies the working principle and structural design of the module, and constructs a college English blended teaching system structure based on MSD fusion. From the evaluation results, it can be seen that the college English blended teaching system based on MSD fusion proposed in this paper can effectively improve the quality of English teaching.