The extraction of typical features of underwater target signals and excellent recognition algorithms are the keys to achieving underwater acoustic target recognition of divers. This paper proposes a feature extraction method for diver signals: frequency-domain multi-sub-band energy (FMSE), aiming to achieve accurate recognition of diver underwater acoustic targets by passive sonar. The impact of the presence or absence of targets, different numbers of targets, different signal-to-noise ratios, and different detection distances on this method was studied based on experimental data under different conditions, such as water pools and lakes. It was found that the FMSE method has the best robustness and performance compared with two other signal feature extraction methods: mel frequency cepstral coefficient filtering and gammatone frequency cepstral coefficient filtering. Combined with the commonly used recognition algorithm of support vector machines, the FMSE method can achieve a comprehensive recognition accuracy of over 94% for frogman underwater acoustic targets. This indicates that the FMSE method is suitable for underwater acoustic recognition of diver targets.

The potential for rotor component shedding in rotating machinery poses significant risks, necessitating the development of an early and precise fault diagnosis technique to prevent catastrophic failures and reduce maintenance costs. This study introduces a data-driven approach to detect rotor component shedding at its inception, thereby enhancing operational safety and minimizing downtime. Utilizing frequency analysis, this research identifies harmonic amplitudes within rotor vibration data as key indicators of impending faults. The methodology employs principal component analysis (PCA) to orthogonalize and reduce the dimensionality of vibration data from rotor sensors, followed by k-fold cross-validation to select a subset of significant features, ensuring the detection algorithm\'s robustness and generalizability. These features are then integrated into a linear discriminant analysis (LDA) model, which serves as the diagnostic engine to predict the probability of rotor component shedding. The efficacy of the approach is demonstrated through its application to 16 industrial compressors and turbines, proving its value in providing timely fault warnings and enhancing operational reliability.

BACKGROUND: The use of machine learning in medical diagnosis and treatment has grown significantly in recent years with the development of computer-aided diagnosis systems, often based on annotated medical radiology images. However, the lack of large annotated image datasets remains a major obstacle, as the annotation process is time-consuming and costly. This study aims to overcome this challenge by proposing an automated method for annotating a large database of medical radiology images based on their semantic similarity.
RESULTS: An automated, unsupervised approach is used to create a large annotated dataset of medical radiology images originating from the Clinical Hospital Centre Rijeka, Croatia. The pipeline is built by data-mining three different types of medical data: images, DICOM metadata and narrative diagnoses. The optimal feature extractors are then integrated into a multimodal representation, which is then clustered to create an automated pipeline for labelling a precursor dataset of 1,337,926 medical images into 50 clusters of visually similar images. The quality of the clusters is assessed by examining their homogeneity and mutual information, taking into account the anatomical region and modality representation.
CONCLUSIONS: The results indicate that fusing the embeddings of all three data sources together provides the best results for the task of unsupervised clustering of large-scale medical data and leads to the most concise clusters. Hence, this work marks the initial step towards building a much larger and more fine-grained annotated dataset of medical radiology images.

The use of anti-inflammatory peptides (AIPs) as an alternative therapeutic approach for inflammatory diseases holds great research significance. Due to the high cost and difficulty in identifying AIPs with experimental methods, the discovery and design of peptides by computational methods before the experimental stage have become promising technology. In this study, we present BertAIP, a bidirectional encoder representation from transformers (BERT)-based method for predicting AIPs directly from their amino acid sequence without using any other information. BertAIP implements a BERT model to extract features of a protein, and uses a fully connected feed-forward network for AIP classification. It was constructed and evaluated using the AIP datasets that were reconstructed from the latest Immune Epitope Database. The experimental results showed that BertAIP achieved an accuracy of 0.751 and a Matthews correlation coefficient of 0.451, which were higher than other commonly used methods. The results of the independent test suggested that BertAIP outperformed the existing AIP predictors. In addition, to enhance the interpretability of BertAIP, we explored and visualized the amino acids that the model considered important for AIP prediction. We believe that the BertAIP proposed herein will be a useful tool for large-scale screening and identifying novel AIPs for drug development and therapeutic research related to inflammatory diseases.

Fine-tuning is an important technique in transfer learning that has achieved significant success in tasks that lack training data. However, as it is difficult to extract effective features for single-source domain fine-tuning when the data distribution difference between the source and the target domain is large, we propose a transfer learning framework based on multi-source domain called adaptive multi-source domain collaborative fine-tuning (AMCF) to address this issue. AMCF utilizes multiple source domain models for collaborative fine-tuning, thereby improving the feature extraction capability of model in the target task. Specifically, AMCF employs an adaptive multi-source domain layer selection strategy to customize appropriate layer fine-tuning schemes for the target task among multiple source domain models, aiming to extract more efficient features. Furthermore, a novel multi-source domain collaborative loss function is designed to facilitate the precise extraction of target data features by each source domain model. Simultaneously, it works towards minimizing the output difference among various source domain models, thereby enhancing the adaptability of the source domain model to the target data. In order to validate the effectiveness of AMCF, it is applied to seven public visual classification datasets commonly used in transfer learning, and compared with the most widely used single-source domain fine-tuning methods. Experimental results demonstrate that, in comparison with the existing fine-tuning methods, our method not only enhances the accuracy of feature extraction in the model but also provides precise layer fine-tuning schemes for the target task, thereby significantly improving the fine-tuning performance.

UNASSIGNED: Dyslexia is a neurological disorder that affects an individual\'s language processing abilities. Early care and intervention can help dyslexic individuals succeed academically and socially. Recent developments in deep learning (DL) approaches motivate researchers to build dyslexia detection models (DDMs). DL approaches facilitate the integration of multi-modality data. However, there are few multi-modality-based DDMs.
UNASSIGNED: In this study, the authors built a DL-based DDM using multi-modality data. A squeeze and excitation (SE) integrated MobileNet V3 model, self-attention mechanisms (SA) based EfficientNet B7 model, and early stopping and SA-based Bi-directional long short-term memory (Bi-LSTM) models were developed to extract features from magnetic resonance imaging (MRI), functional MRI, and electroencephalography (EEG) data. In addition, the authors fine-tuned the LightGBM model using the Hyperband optimization technique to detect dyslexia using the extracted features. Three datasets containing FMRI, MRI, and EEG data were used to evaluate the performance of the proposed DDM.
UNASSIGNED: The findings supported the significance of the proposed DDM in detecting dyslexia with limited computational resources. The proposed model outperformed the existing DDMs by producing an optimal accuracy of 98.9%, 98.6%, and 98.8% for the FMRI, MRI, and EEG datasets, respectively. Healthcare centers and educational institutions can benefit from the proposed model to identify dyslexia in the initial stages. The interpretability of the proposed model can be improved by integrating vision transformers-based feature extraction.

The purpose of this study is to put forward a feature extraction and pattern recognition method for the flow noise signal of natural gas pipelines in view of the complex situation brought by the rapid development and expansion of urban natural gas infrastructure in China, especially in the case that there are active and abandoned pipelines, metal and nonmetal pipelines, and natural gas, water and power pipelines coexist in the underground of the city. Because the underground situation is unknown, gas leakage incidents caused by natural gas pipeline rupture occur from time to time, posing a threat to personal safety. Therefore, the motivation of this study is to provide a feasible method to accelerate the aging, renewal and transformation of urban natural gas pipelines to ensure the safe operation of urban natural gas pipeline network and promote the high-quality development of urban economy. Through the combination of experimental test and numerical simulation, this study establishes a database of urban natural gas pipeline flow noise signals, and uses principal component analysis (PCA) to extract the characteristics of flow noise signals, and develops a mathematical model for feature extraction. Then, a classification and recognition model based on backpropagation neural network (BPNN) is constructed, which realizes the detection and recognition of convective noise signals. The research results show that the theoretical method based on acoustic feature analysis provides guidance for the orderly and safe construction of urban natural gas pipeline network and ensures its safe operation. The research conclusion shows that through the simulation analysis of 75 groups of gas pipeline flow noise under different working conditions. Combined with the experimental verification of ground flow noise signals, the feature extraction and pattern recognition method proposed in this study has a recognition accuracy of up to 97% under strong noise background, which confirms the accuracy of numerical simulation and provides theoretical basis and technical support for the detection and recognition of urban gas pipeline flow noise.

To improve the classification and regression performance of the total volatile basic nitrogen (TVB-N) and acid value (AV) of different freshness fish meal samples detected by a metal-oxide semiconductor electronic nose (MOS e-nose), 402 original features, 62 manually extracted features, manually extracted and selected features by the RFRFE method, and the features extracted by the long short-term memory (LSTM) network were used as inputs to identify the freshness. The classification performance of the freshness grades and the estimation performance of the TVB-N and AV values of fish meal with different freshness were compared. According to the sensor response curve, preprocessing and feature extraction steps were first applied to the original data. Then, five classification algorithms and four regression algorithms were used for modeling. The results showed that a total of 30 features were extracted using the LSTM network, and the number of extracted features was significantly reduced. In the classification, the highest accuracy rate of 95.4% was obtained using the support vector machine method. In the regression, the least squares support vector regression method obtained the best root mean square error (RMSE). The coefficient of determination (R2), RMSE, and relative standard deviation (RSD) between the predicted value of TVBN and the actual value were 0.963, 11.01, and 7.9%, respectively. The R2, RMSE, and RSD between the predicted value of AV and the actual value were 0.972, 0.170, and 6.05%, respectively. The LSTM feature extraction method provided a new method and reference for feature extraction using an E-nose to identify other animal-derived material samples.

The panoramic stereo video has brought a new visual experience for the audience with its immersion and stereo effect. In panoramic stereo video, the face is an important element. However, the face image in panoramic stereo video has varying degrees of deformation. This brings new challenges to face recognition. Therefore, this paper proposes a face recognition model DCM2Net (Deformable Convolution MobileFaceNet) for panoramic stereo video. The model mainly integrates the feature information between channels during feature fusion, redistributes the information between channels in the deeper part of the network, and fully uses the information between different channels for feature extraction. This paper also built a panoramic stereo video live system, using the DCM2Net model to recognize the face in panoramic stereo video, and the recognition results are displayed in the video. After experiments on different datasets, the results show that our model has better results on popular datasets and panoramic datasets.

In this study, the goal was to develop a method for detecting and classifying organophosphorus pesticides (OPPs) in bodies of water. Sixty-five samples with different concentrations were prepared for each of the organophosphorus pesticides, namely chlorpyrifos, acephate, parathion-methyl, trichlorphon, dichlorvos, profenofos, malathion, dimethoate, fenthion, and phoxim, respectively. Firstly, the spectral data of all the samples was obtained using a UV-visible spectrometer. Secondly, five preprocessing methods, six manifold learning methods, and five machine learning algorithms were utilized to build detection models for identifying OPPs in water bodies. The findings indicate that the accuracy of machine learning models trained on data preprocessed using convolutional smoothing + first-order derivatives (SG + FD) outperforms that of models trained on data preprocessed using other methods. The backpropagation neural network (BPNN) model exhibited the highest accuracy rate at 99.95%, followed by the support vector machine (SVM) and convolutional neural network (CNN) models, both at 99.92%. The extreme learning machine (ELM) and K-nearest neighbors (KNN) models demonstrated accuracy rates of 99.84% and 99.81%, respectively. Following the application of a manifold learning algorithm to the full-wavelength data set for the purpose of dimensionality reduction, the data was then visualized in the first three dimensions. The results demonstrate that the t-distributed domain embedding (t-SNE) algorithm is superior, exhibiting dense clustering of similar clusters and clear classification of dissimilar ones. SG + FD-t-SNE-SVM ranks highest among the feature extraction models in terms of performance. The feature extraction dimension was set to 4, and the average classification accuracy was 99.98%, which slightly improved the prediction performance over the full-wavelength model. As shown in this study, the ultraviolet-visible (UV-visible) spectroscopy system combined with the t-SNE and SVM algorithms can effectively identify and classify OPPs in waterbodies.