Second-order Features Research Articles

SSDStacked-BLS with Extended Depth and Width: Infrared Fault Diagnosis of Rolling Bearings under Dual Feature Selection

In fault diagnosis, broad learning systems (BLS) have been applied in recent years. However, the best fault diagnosis cannot be guaranteed by width node extension alone, so a stacked broad learning system (stacked BLS) was proposed. Most of the methods for choosing the number of depth layers used optimization algorithms that tend to increase computation time. In addition, the data under single feature selection are not sufficiently representative, and effective features are easily lost. To solve these problems, this article proposes an infrared fault diagnosis model for rolling bearings based on integration of principal component analysis and singular value decomposition (IPS) and the stacked BLS with self-selected depth model (SSDStacked-BLS). First, 72 second-order statistical features are extracted from the pre-processed infrared images of rolling bearings. Next, feature selection is performed using IPS. he IPS feature selection module consists of principal component analysis (PCA) and singular value decomposition (SVD). The feature selection is performed by PCA and SVD separately, which are then stitched together to form a new feature. This ensures a comprehensive coverage of infrared image features. Finally, the acquired features are input into SSDStacked-BLS. This model establishes a data storage group for the residual training characteristics of stacked BLS, adding one block at a time. The accuracy rate of each newly added block is output and saved to the data storage group. If the diagnostic rate fails to increase three consecutive times, the block stacking is stopped and the results are output. IPS-SSDStacked-BLS achieved an accuracy of 0.9667 in 0.1775 s. This is almost five times faster than stacked BLS optimized using the grid search method. Compared with the original BLS, its accuracy was 0.0445 higher and the time was approximated. Compared with IPS-SVM, IPS-RF, IPS-1DCNN and 2DCNN, IPS-SSDStacked-BLS was more advantageous in terms of accuracy and time consumption.

Graph ‘texture’ features as novel metrics that can summarize complex biological graphs

Objective. Image texture features, such as those derived by Haralick et al, are a powerful metric for image classification and are used across fields including cancer research. Our aim is to demonstrate how analogous texture features can be derived for graphs and networks. We also aim to illustrate how these new metrics summarize graphs, may aid comparative graph studies, may help classify biological graphs, and might assist in detecting dysregulation in cancer. Approach. We generate the first analogies of image texture for graphs and networks. Co-occurrence matrices for graphs are generated by summing over all pairs of neighboring nodes in the graph. We generate metrics for fitness landscapes, gene co-expression and regulatory networks, and protein interaction networks. To assess metric sensitivity we varied discretization parameters and noise. To examine these metrics in the cancer context we compare metrics for both simulated and publicly available experimental gene expression and build random forest classifiers for cancer cell lineage. Main results. Our novel graph ‘texture’ features are shown to be informative of graph structure and node label distributions. The metrics are sensitive to discretization parameters and noise in node labels. We demonstrate that graph texture features vary across different biological graph topologies and node labelings. We show how our texture metrics can be used to classify cell line expression by lineage, demonstrating classifiers with 82% and 89% accuracy. Significance. New metrics provide opportunities for better comparative analyzes and new models for classification. Our texture features are novel second-order graph features for networks or graphs with ordered node labels. In the complex cancer informatics setting, evolutionary analyses and drug response prediction are two examples where new network science approaches like this may prove fruitful.

Identification of Radiomic Signatures in Brain MRI Sequences T1 and T2 That Differentiate Tumor Regions of Midline Gliomas with H3.3K27M Mutation.

Radiomics refers to the acquisition of traces of quantitative features that are usually non-perceptible to human vision and are obtained from different imaging techniques and subsequently transformed into high-dimensional data. Diffuse midline gliomas (DMG) represent approximately 20% of pediatric CNS tumors, with a median survival of less than one year after diagnosis. We aimed to identify which radiomics can discriminate DMG tumor regions (viable tumor and peritumoral edema) from equivalent midline normal tissue (EMNT) in patients with the positive H3.F3K27M mutation, which is associated with a worse prognosis. This was a retrospective study. From a database of 126 DMG patients (children, adolescents, and young adults), only 12 had H3.3K27M mutation and available brain magnetic resonance DICOM file. The MRI T1 post-gadolinium and T2 sequences were uploaded to LIFEx software to post-process and extract radiomic features. Statistical analysis included normal distribution tests and the Mann-Whitney U test performed using IBM SPSS® (Version 27.0.0.1, International Business Machines Corp., Armonk, NY, USA), considering a significant statistical p-value ≤ 0.05. EMNT vs. Tumor: From the T1 sequence 10 radiomics were identified, and 14 radiomics from the T2 sequence, but only one radiomic identified viable tumors in both sequences (p < 0.05) (DISCRETIZED_Q1). Peritumoral edema vs. EMNT: From the T1 sequence, five radiomics were identified, and four radiomics from the T2 sequence. However, four radiomics could discriminate peritumoral edema in both sequences (p < 0.05) (CONVENTIONAL_Kurtosis, CONVENTIONAL_ExcessKurtosis, DISCRETIZED_Kurtosis, and DISCRETIZED_ExcessKurtosis). There were no radiomics useful for distinguishing tumor tissue from peritumoral edema in both sequences. Less than 5% of the radiomic characteristics identified tumor regions of medical-clinical interest in T1 and T2 sequences of conventional magnetic resonance imaging. The first-order and second-order radiomic features suggest support to investigators and clinicians for careful evaluation for diagnosis, patient classification, and multimodality cancer treatment planning.

FgbCNN: A unified bilinear architecture for learning a fine-grained feature representation in facial expression recognition

A facial expression recognition system has been proposed in this paper. The challenges of the facial expression recognition system lie due to low intra-class variance within a class of negative emotions, such as anger, disgust, and fear. A conventional Convolution Neural Network (CNN) model may extract discriminative features and has shown outstanding performance in various computer vision tasks. However, it is unable to extract the second-order feature information that demonstrates the interaction between features in the image of wild-environment datasets. This paper presents a novel method to solve the facial expression recognition problem by addressing several limitations concerning emotion recognition problems. A deep learning-based bilinear convolutional neural network framework has been proposed, termed an Fine-Grained Bilinear CNN (FgbCNN) model that consists of two branches with optimized CNN along with a normalization layer composed of batch-normalization, square-root normalization, L2-normalization, and drop-out layers. Here, local and holistic features have been aggregated using a dot-product layer to extract more discriminant features. Finally, experimenting with two wild-environments (SFEW 1.0 and SFEW 2.0) and two lab-controlled datasets (KDEF and CK+), it has been observed that the proposed model can minimize the intra-class variances and has attained outstanding performance compared to other state-of-the-art methods.

MRI-Based Radiomics in Bladder Cancer: A Systematic Review and Radiomics Quality Score Assessment.

(1): Background: With the recent introduction of vesical imaging reporting and data system (VI-RADS), magnetic resonance imaging (MRI) has become the main imaging method used for the preoperative local staging of bladder cancer (BCa). However, the VI-RADS score is subject to interobserver variability and cannot provide information about tumor cellularity. These limitations may be overcome by using a quantitative approach, such as the new emerging domain of radiomics. (2) Aim: To systematically review published studies on the use of MRI-based radiomics in bladder cancer. (3) Materials and Methods: We performed literature research using the PubMed MEDLINE, Scopus, and Web of Science databases using PRISMA principles. A total of 1092 papers that addressed the use of radiomics for BC staging, grading, and treatment response were retrieved using the keywords "bladder cancer", "magnetic resonance imaging", "radiomics", and "textural analysis". (4) Results: 26 papers met the eligibility criteria and were included in the final review. The principal applications of radiomics were preoperative tumor staging (n = 13), preoperative prediction of tumor grade or molecular correlates (n = 9), and prediction of prognosis/response to neoadjuvant therapy (n = 4). Most of the developed radiomics models included second-order features mainly derived from filtered images. These models were validated in 16 studies. The average radiomics quality score was 11.7, ranging between 8.33% and 52.77%. (5) Conclusions: MRI-based radiomics holds promise as a quantitative imaging biomarker of BCa characterization and prognosis. However, there is still need for improving the standardization of image preprocessing, feature extraction, and external validation before applying radiomics models in the clinical setting.

Computer-aided recognition and assessment of a porous bioelastomer in ultrasound images for regenerative medicine applications

It is difficult to use a single edge operator in image processing to extract continuous and accurate contours of a porous bioelastomer due to the fuzzy boundary and complex background in ultrasound images. To solve this problem, this paper proposes a joint algorithm for bioelastomer contour detection and a texture feature extraction method for monitoring the degradation performance of bioelastomers. First, the mean-shift clustering method is utilized to obtain the clustering feature information of bioelastomers and native tissue from manually segmented images, and this information is used as the initial information in the image binarization algorithm for image partitioning. Second, Otsu's thresholding method and mathematical morphology are applied in the process of image binarization. Finally, the Canny edge detector is employed to extract the complete bioelastomers contour from the binary image. To verify the robustness of the proposed joint algorithm, the results using the proposed joint algorithm, where mean-shift clustering is replaced with k-means clustering are also obtained. The proposed joint algorithm based on mean-shift clustering outperforms the joint algorithm based on k-means clustering, as well as algorithms that directly apply the Canny, Sobel and Laplacian methods. Texture feature extraction is based on the computer-aided recognition of bioelastomers. The region of interest (ROI) is set in the scaffold region, and the first-order statistical features and second-order statistical features of the greyscale values of the ROI are extracted and analysed. The proposed joint algorithm can not only extract ideal bioelastomers contours from ultrasound images but also provide valuable feedback on the degradation behaviour of bioelastomers at implant sites.

RFR-GA-BLS: A Feature Selection and Parameter Optimization Method for Fault Diagnosis of Rolling Bearing Using Infrared Images

To overcome the problems of low machine learning fault diagnosis rate and long consumption time of deep learning in rolling bearing fault diagnosis, an RFR-GA-BLS model is proposed. The model is validated by infrared images of rolling bearings to find the most representative features, the most suitable parameters and the best diagnostic rate. Based on the pre-processed infrared thermal images of the faulty bearing, 72 second-order statistical features were obtained as information for fault diagnosis. RFR considered the robustness of the features, and new sequences were obtained. BLS was optimized by GA for fault diagnosis. New sequence features were added to the model sequentially, one at a time. After satisfying the model conditions, the most appropriate number of features was selected as the first 20. The search results for the number of feature nodes, the number of feature node windows and the number of enhancement nodes for the BLS were 24, 19 and 544, respectively, and the fault diagnosis rate of 98.8889% was achieved. According to a comparison with CFR-GA-BLS, BLS, PSO-BLS and Grdy-BLS, our proposed model is more advantageous in the search for the best performance. The fault diagnosis accuracy is higher compared to SVM and RF. The speed of our proposed model is 207 times faster than 1DCNN and 10,147 times faster than 2DCNN.

A method of knowledge distillation based on feature fusion and attention mechanism for complex traffic scenes

Object detectors based on deep learning can run smoothly on a terminal device in complex traffic scenes, and the model compression method has become a research hotspot. Considering student network single learning in the knowledge distillation algorithm, the dependence on loss function design leads to parameter sensitivity and other problems, we propose a new knowledge distillation method with second-order term attention mechanisms and feature fusion of adjacent layers. First, we build a knowledge distillation framework based on YOLOv5 and propose a new attention mechanism in the teacher network backbone to extract the hot map. Then, we combine the hot map features with the next level features through the fusion module. By fusing the useful information of the low convolution layer and the feature map of the high convolution layer to help the student network obtain the final prediction map. Finally, to improve the accuracy of small objects, we add a 160 × 160 detection head and use a transformer encoder block module to replace the convolution network of the head. Sufficient experimental results show that our method achieves state-of-the-art performance. The speed and number of parameters remain unchanged, but the average detection accuracy is 97.4% on the KITTI test set. On the Cityscapes test set, the average detection accuracy reaches 92.7%.

Role of MRI radiomics for the prediction of MYCN amplification in neuroblastomas.

We evaluate MR radiomics and develop machine learning-based classifiers to predict MYCN amplification in neuroblastomas. A total of 120 patients with neuroblastomas and baseline MR imaging examination available were identified of whom 74 (mean age ± standard deviation [SD] of 6years and 2months ± 4years and 9months; 43 females and 31 males, 14 MYCN amplified) underwent imaging at our institution. This was therefore used to develop radiomics models. The model was tested in a cohort of children with the same diagnosis but imaged elsewhere (n = 46, mean age ± SD: 5years 11months ± 3years 9months, 26 females and 14 MYCN amplified). Whole tumour volumes of interest were adopted to extract first-order histogram and second-order radiomics features. Interclass correlation coefficient and maximum relevance and minimum redundancy algorithm were applied for feature selection. Logistic regression, support vector machine, and random forest were employed as the classifiers. Receiver operating characteristic (ROC) analysis was performed to evaluate the diagnostic accuracy of the classifiers on the external test set. The logistic regression model and the random forest both showed an AUC of 0.75. The support vector machine classifier obtained an AUC of 0.78 on the test set with a sensitivity of 64% and a specificity of 72%. The study provides preliminary retrospective evidence demonstrating the feasibility of MRI radiomics in predicting MYCN amplification in neuroblastomas. Future studies are needed to explore the correlation between other imaging features and genetic markers and to develop multiclass predictive models. • MYCN amplification in neuroblastomas is an important determinant of disease prognosis. • Radiomics analysis of pre-treatment MR examinations can be used to predict MYCN amplification in neuroblastomas. • Radiomics machine learning models showed good generalisability to external test set, demonstrating reproducibility of the computational models.

Modeling of wave propagation in polycrystalline ice with hierarchical density gradients

Polycrystalline solids are composed of many small grains of varying sizes and crystallographic orientations. An elastic wave that propagates through such a material experiences distortion and attenuation. While the influence on propagation in random configurations can be captured with conventional statistical descriptors, the role of second-order features such as the hierarchical gradient in material properties has not been explored. In this paper, we optimize a numerical strategy based on Finite Elements and Local Max-Entropy approximants to characterize the role of grain density gradients on ultrasonic attenuation. We focus on ice as a model for mesoscale ordered configurations due to its relevance to the emerging technology of cryoultrasonics. Our simulations in one- and two-dimensional settings indicate that second-order descriptors are required to predict attenuation in polycrystalline ice. Furthermore, we define a novel parameter, based on the standard deviation of the speed of sound gradient distribution, which shows a quadratic relationship with the ultrasonic attenuation. The model results can be understood as a phase diagram for the design of metamaterials with specific ultrasonic scattering properties.

Second-Order Statistics of Self-Splitting Structured Beams in Oceanic Turbulence

Free-space optical communication is restricted by random media-stimulated beam degradation. However, partially coherent structured beams modulated by the coherence structure can potentially mitigate this negative effect. By employing the extended Huygens–Fresnel integral, we provide an examination of the second-order statistical features of a common type of partly coherent structured beams, self-splitting structured beams, in a turbulent ocean. The implications of turbulence parameters relating to the ocean and beginning beam parameters corresponding to the progression of such beam propagation attributes are fully investigated. Our numerical outcomes show that, for turbulence with a low-dissipation kinetic energy rate per unit mass of fluid, small Kolmogorov inner scale, large relative strength of temperature to salinity undulations, and large dissipation rate of mean-square temperature has a greater negative effect on the structured beams. In addition, we suggest an effective approach, enhancing the order of the beam and reducing the coherence length of the beams, to lower the oceanic turbulence-induced negative effects, and thus have future extensive possibilities in free-space optical communication.

Combining CNN Features with Voting Classifiers for Optimizing Performance of Brain Tumor Classification.

Brain tumors and other nervous system cancers are among the top ten leading fatal diseases. The effective treatment of brain tumors depends on their early detection. This research work makes use of 13 features with a voting classifier that combines logistic regression with stochastic gradient descent using features extracted by deep convolutional layers for the efficient classification of tumorous victims from the normal. From the first and second-order brain tumor features, deep convolutional features are extracted for model training. Using deep convolutional features helps to increase the precision of tumor and non-tumor patient classification. The proposed voting classifier along with convoluted features produces results that show the highest accuracy of 99.9%. Compared to cutting-edge methods, the proposed approach has demonstrated improved accuracy.

Rock Thin Section Image Identification Based on Convolutional Neural Networks of Adaptive and Second-Order Pooling Methods

In order to enhance the ability to represent rock feature information and finally improve the rock identification performance of convolution neural networks (CNN), a new pooling mode was proposed in this paper. According to whether the pooling object was the last convolution layer, it divided pooling layers into the sampling pooling layer and the classification pooling layer. The adaptive pooling method was used in the sampling pooling layer. The pooling kernels adaptively adjusted were designed for each feature map. The second-order pooling method was used by the classification pooling layer. The second-order feature information based on outer products was extracted from the feature pair. The changing process of the two methods in forward and back propagation was deduced. Then, they were embedded into CNN to build a rock thin section image identification model (ASOPCNN). The experiment was conducted on the image set containing 5998 rock thin section images of six rock types. The CNN models using max pooling, average pooling and stochastic pooling were set for comparison. In the results, the ASOPCNN has the highest identification accuracy of 89.08% on the test set. Its indexes are superior to the other three models in precision, recall, F1 score and AUC values. The results reveal that the adaptive and second-order pooling methods are more suitable for CNN model, and CNN based on them could be a reliable model for rock identification.

Correlation feature distribution matching for fault diagnosis of machines

• A correlated feature distribution matching (CFDM) is proposed for fault diagnosis • Aligning the features of two domains can improve the feature adaptation performance • By matching the weights of two distributions, the distribution shift can be reduced • CFDM accurately diagnoses unlabeled samples using labeled samples with similar status • Three bearing experiments are performed to verify the validity of the proposed method The variation of working conditions makes the probability distribution of the source domain data and the target domain data differ greatly, which leads to the performance degradation of conventional fault diagnosis methods. Transfer learning is an effective tool to deal with this issue. However, existing methods still have shortcomings since they generally treat the two distributions equally and thus cannot effectively adjust the relative importance of the two distributions. What's more, they focus more on the probability distribution but neglect to align the features of the two domains. To this end, this study proposes a framework called correlated feature distribution matching (CFDM) to efficiently achieve cross-domain fault diagnosis. Firstly, CFDM finds the correlation information between the source and target domains by the correlation feature matching, and then performs second-order feature alignment for the two domains, thus reducing the difficulty of feature adaptation. This can not only reduce the original feature distance between two domains, but also effectively avoid the feature distortion caused by the loss of the original key information in the feature transformation, so as to obtain the correlation features. Secondly, CFDM takes into account both the marginal distribution and the conditional distribution, and dynamically adjusts the relative importance of the two distributions of the correlation features through the feature dynamic adaptation. This can precisely match the weights of the two distributions and further reduce the distribution difference between the two domains. Finally, the validity and reliability of the proposed CFDM are verified on three bearing test rigs.

Diagnostic Performance of Selected MRI-Derived Radiomics Able to Discriminate Progression-Free and Overall Survival in Patients with Midline Glioma and the H3F3AK27M Mutation.

Radiomics refers to a recent area of knowledge that studies features extracted from different imaging techniques and subsequently transformed into high-dimensional data that can be associated with biological events. Diffuse midline gliomas (DMG) are one of the most devastating types of cancer, with a median survival of approximately 11 months after diagnosis and 4-5 months after radiological and clinical progression. A retrospective study. From a database of 91 patients with DMG, only 12 had the H3.3K27M mutation and brain MRI DICOM files available. Radiomic features were extracted from MRI T1 and T2 sequences using LIFEx software. Statistical analysis included normal distribution tests and the Mann-Whitney U test, ROC analysis, and calculation of cut-off values. A total of 5760 radiomic values were included in the analyses. AUROC demonstrated 13 radiomics with statistical significance for progression-free survival (PFS) and overall survival (OS). Diagnostic performance tests showed nine radiomics with specificity for PFS above 90% and one with a sensitivity of 97.2%. For OS, 3 out of 4 radiomics demonstrated between 80 and 90% sensitivity. Several radiomic features demonstrated statistical significance and have the potential to further aid DMG diagnostic assessment non-invasively. The most significant radiomics were first- and second-order features with GLCM texture profile, GLZLM_GLNU, and NGLDM_Contrast.

Outcome Prediction of Spontaneous Supratentorial Intracerebral Hemorrhage after Surgical Treatment Based on Non-Contrast Computed Tomography: A Multicenter Study.

This study aims to explore the value of a machine learning (ML) model based on radiomics features and clinical features in predicting the outcome of spontaneous supratentorial intracerebral hemorrhage (sICH) 90 days after surgery. A total of 348 patients with sICH underwent craniotomy evacuation of hematoma from three medical centers. One hundred and eight radiomics features were extracted from sICH lesions on baseline CT. Radiomics features were screened using 12 feature selection algorithms. Clinical features included age, gender, admission Glasgow Coma Scale (GCS), intraventricular hemorrhage (IVH), midline shift (MLS), and deep ICH. Nine ML models were constructed based on clinical feature, and clinical features + radiomics features, respectively. Grid search was performed on different combinations of feature selection and ML model for parameter tuning. The averaged receiver operating characteristics (ROC) area under curve (AUC) was calculated and the model with the largest AUC was selected. It was then tested using multicenter data. The combination of lasso regression feature selection and logistic regression model based on clinical features + radiomics features had the best performance (AUC: 0.87). The best model predicted an AUC of 0.85 (95%CI, 0.75-0.94) on the internal test set and 0.81 (95%CI, 0.64-0.99) and 0.83 (95%CI, 0.68-0.97) on the two external test sets, respectively. Twenty-two radiomics features were selected by lasso regression. The second-order feature gray level non-uniformity normalized was the most important radiomics feature. Age is the feature with the greatest contribution to prediction. The combination of clinical features and radiomics features using logistic regression models can improve the outcome prediction of patients with sICH 90 days after surgery.

Prediction of Wilms' Tumor Susceptibility to Preoperative Chemotherapy Using a Novel Computer-Aided Prediction System.

Wilms' tumor, the most prevalent renal tumor in children, is known for its aggressive prognosis and recurrence. Treatment of Wilms' tumor is multimodal, including surgery, chemotherapy, and occasionally, radiation therapy. Preoperative chemotherapy is used routinely in European studies and in select indications in North American trials. The objective of this study was to build a novel computer-aided prediction system for preoperative chemotherapy response in Wilms' tumors. A total of 63 patients (age range: 6 months-14 years) were included in this study, after receiving their guardians' informed consent. We incorporated contrast-enhanced computed tomography imaging to extract the texture, shape, and functionality-based features from Wilms' tumors before chemotherapy. The proposed system consists of six steps: (i) delineate the tumors' images across the three contrast phases; (ii) characterize the texture of the tumors using first- and second-order textural features; (iii) extract the shape features by applying a parametric spherical harmonics model, sphericity, and elongation; (iv) capture the intensity changes across the contrast phases to describe the tumors' functionality; (v) apply features fusion based on the extracted features; and (vi) determine the final prediction as responsive or non-responsive via a tuned support vector machine classifier. The system achieved an overall accuracy of 95.24%, with 95.65% sensitivity and 94.12% specificity. Using the support vector machine along with the integrated features led to superior results compared with other classification models. This study integrates novel imaging markers with a machine learning classification model to make early predictions about how a Wilms' tumor will respond to preoperative chemotherapy. This can lead to personalized management plans for Wilms' tumors.

An attention-based bilinear feature extraction mechanism for fine-grained laser welding molten pool/keyhole defect recognition

Vision-based molten pool/keyhole (MPK) defect recognition is essential for online monitoring of laser welding quality. However, the visual differences corresponding to different MPK states are extremely small, which especially brings great challenges to the recognition of defect subclasses. In this paper, an attention-based bilinear feature extraction mechanism is developed for recognizing fine-grained laser welding MPK defects. First, first-order linear features of MPK image are extracted through a skeleton network. Second, spatial attention and channel attention are designed to extract first-order functional features from the first-order linear features of the MPK image, and the attention mechanism explicitly models the network's ability to learn discriminative location and attribute features. Third, second-order bilinear features are extracted from first-order functional features of the MPK image by a bilinear confluence operation and used for fine-grained MPK image recognition. Finally, three laser welding MPK image datasets with different fine-grained degrees are constructed, and the proposed method is comprehensively tested for ablation and comparison. The results show that the proposed method has an accuracy of 97.45 % for the recognition of 9 types of laser welding MPK defects.

Internal Leakage Detection in Hydraulic Pump Using Model-Agnostic Feature Ranking and Ensemble Classifiers

Abstract Hydraulic pumps are key drivers of fluid power-based machines and demand high reliability during operation. Internal leakage is a key performance deteriorating fault that reduces pump’s efficiency and limits its predictability and reliability. Thus, this article presents a methodology for detecting internal leakage in hydraulic pumps using an unbalanced dataset of its drive motor’s electrical power signals. Refined composite multiscale dispersion and fuzzy entropies along with three statistical indicators are extracted and followed by second-order polynomial-based features. These features are normalized and visualized using partial dependence plot (PDP) and individual conditional expectation (ICE). Subsequently, ten machine learning classifiers are trained using four features, and their statistical hypothesis test is performed using a 5 × 2 paired t-test cross-validation for p &amp;lt; 0.05. Subsequently, top four performing classifiers are optimized using grid and random search hyperparameter optimization techniques. Due to slight difference in their accuracies, an ensemble of three best-performing algorithms is trained using the majority voting classifiers (MaVCs) for three splitting ratios (80:20, 70:30, and 60:40). It is demonstrated that MaVC achieves the highest leakage detection accuracy of 90.91%.

Scene Classification Based on Heterogeneous Features of Multi-Source Data

Remote sensing scene classification is quite important in earth observation and other fields. Previous research has found that most of the existing models are based on deep learning models. However, the classification accuracy of the deep learning model is difficult to break through due to the challenges of difficulty distinguishing the socio-economic attributes of scenes, high interclass similarity, and large intraclass differences. To tackle the challenges, we propose a novel scene classification model that integrates heterogeneous features of multi-source data. Firstly, a multi-granularity feature learning module is designed, which can conduct uniform grid sampling of images to learn multi-granularity features. In this module, in addition to the features of our previous research, we also supplemented the socio-economic semantic features of the scene, and attention-based pooling is introduced to achieve different levels of representation of images. Then, to reduce the dimension of the feature, we adopt the feature-level fusion method. Next, the maxout-based module is designed to fuse the features of different granularity and extract the most distinguishing second-order latent ontology essence features. The weighted adaptive fusion method is used to fuse all the features. Finally, the Lie Group Fisher algorithm is used for scene classification. Extensive experimentation and evaluations show that our proposed model can find better solutions to the above challenges.