One-dimensional Convolutional Neural Network Research Articles

An integrated deep neural network model combining 1D CNN and LSTM for structural health monitoring utilizing multisensor time-series data

Introducing deep learning algorithms into the field of structural health monitoring (SHM) has contributed to the automatic extraction of damage-sensitive features, but the type and architecture of these algorithms are still in dispute. This paper proposes a hybrid deep learning framework entitled time-distributed one-dimensional convolutional neural network (1D CNN) long short-term memory (LSTM) model, which utilizes raw multisensor time histories to detect structural damages. Using a sliding window that moves along the temporal dimension, the multisensor data are first segmented into subsequences. The 1D CNN layers are simultaneously applied to each subsequence to extract damage-sensitive features from row data samples. These features are then fed into the LSTM layers to extract temporal features between subsequences. As the final step, these extracted features are classified using fully connected layers. In order to assess the performance of this model, a numerical model of a high-rise frame with nonlinear members is used. This hybrid model is assumed to identify the location of damages to this frame. In order to assess the proposed model with a real-world structure, a well-known benchmark building is employed to identify damage patterns by this deep hybrid neural network. A set of metrics related to the performance of the model is measured and evaluated. It is found that the model has an average accuracy of above 96.6% in localizing damage in the numerical structure and above 99.6% in detecting each damage pattern in the experimental building. The results indicate that the proposed model can be applied effectively to the SHM of different structural systems with different damage patterns.

Assessing the reliability of a physical-based model and a convolutional neural network in an ungauged watershed for daily streamflow calculation: a case study in southern Portugal

The main goal of this study was to estimate inflows to the Maranhão reservoir, southern Portugal, using two distinct modeling approaches: a one-dimensional convolutional neural network (1D-CNN) model and a physically based model. The 1D-CNN was previously trained, validated, and tested in a sub-basin of the study area where observed streamflow values were available. The trained model was here subject to an improvement and applied to the entire watershed by replacing the forcing variables (accumulated and delayed precipitation) to make them correspond to the values of the entire watershed. The same way, the physically based MOHID-Land model was calibrated and validated for the same sub-basin, and the calibrated parameters were then applied to the entire watershed. Inflow values estimated by both models were validated considering a mass balance at the reservoir. The 1D-CNN model demonstrated a better performance in simulating daily values, peak flows, and the wet period. The MOHID-Land model showed a better performance in estimating streamflow values during dry periods and for a monthly analysis. Hence, results show the adequateness of both modeling solutions for integrating a decision support system aimed at supporting decision-makers in the management of water availability in an area subjected to increasing scarcity.

Multi-perspective interpretation for one-dimensional conventional neural network model to identify iron-bearing waste material

Identification of iron-bearing waste material (IWM) is challenging, as the physical and chemical properties of IWMs vary greatly with sources and lacking of common markers. Here, we proposed an interpretable method based on a one-dimensional convolutional neural network (1D CNN) and laser-induced breakdown spectroscopy (LIBS) that can recognize IWM. The proposed 1D CNN model obtained a test accuracy of 99.08%±1.37% with five-time repetition and showed sensitivities and specificities per class within the range of 98%-99%. To shed light into the black-box nature of neural network, a multi-perspective interpretation of CNN algorithm towards the analysis of spectroscopy was performed to provide a more complete view of feature importance, intermediate outputs and decision process. As for identifying IWM, the most critical elements captured by the CNN using a visualization algorithm (class activation map) were Na, Mn, Al, Ca, Fe, Mg, K, Cr, Zn, F and Cl. In the intermediate outputs, it was found that the kernels in the same convolution extract different local informative variables and contribute to learning high-level features. In the final decision process, the CNN was proved to accumulate more valuable information over the increased convolutions using another visualization algorithm (t-SNE) and the output was calculated through mathematical inference. This is the first study in which CNN-assisted LIBS spectroscopy was used to identify IWM while providing a multi-perspective interpretable method for CNN algorithm, which can lay a foundation for widening the application of CNN-based spectroscopic techniques in waste identification.

Prediction of Sino-Korean cultural product trade based on one-dimensional convolutional neural network and support vector machine

This study explores a predictive approach using a combination of a one-dimensional convolutional neural network and support vector machine to enhance the management of cultural product trade between China and South Korea, addressing the trade deficit challenge. The methodology involves the collection and categorization of diverse data related to the trade of cultural products between the two countries, identifying data mining directions. The research incorporates the design of association rule functions to identify viable data sources, and employs a hybrid data clustering algorithm integrating technology and spectral clustering to cluster available data. The features extracted from the data mining process are utilized as learning samples for trade prediction. Both a one-dimensional convolutional neural network and support vector machine are employed to model and predict cultural product trade between China and South Korea. Experimental results demonstrate the method’s accuracy in predicting trade situations under parameterized conditions. Throughout the prediction process, credibility measurement values and controllable correlation degrees consistently exceed 19 and 12.5, respectively, while uncertainty discrimination degrees and error coefficients remain below 12 and 6.

Rapid quantitative analysis of cotton-polyester blended fabrics using near-infrared spectroscopy combined with CNN-LSTM

Accurate and rapid quantitative analysis of fiber content is important for blended fabric quality control and supervision. In this paper, a deep learning method based on an improved hybrid network of convolutional neural network (CNN) and long short-term memory (LSTM) to fast and accurate determine cotton content of cotton-polyester blended fabrics using near-infrared (NIR) spectroscopy. The cotton-polyester blended fabric samples were partitioned into training set, validation set and testing set in the ratio of 6:2:2 to develop the quantitative model of cotton content of blended fabric. Firstly, a one-dimensional CNN architecture is designed and modeled to obtain the CNN parameters and to evaluate the spectral preprocessing methods. Then, the CNN structure is modified and improved using LSTM to obtain a CNN-LSTM structure. The root mean square error (RMSE) of the validation set obtained by CNN model is 0.83%, which is 34.03% lower than that of the traditional partial least squares (PLS) model, and the RMSE of the testing set obtained by CNN model is 0.88%, which is 25.88% lower than that of the PLS model. The RMSE of the validation set obtained by CNN-LSTM model is 0.65%, which is 47.85% lower than that of the PLS model, and the RMSE of the testing set obtained by CNN-LSTM model is 0.75%, which is 36.57% lower than that of the PLS model. Compared with PLS and CNN models, the CNN-LSTM model can more accurately analyze the cotton content proportion of cotton-polyester blended fabrics, which provides a new fast and non-invasive method for the quantitative detection of fabrics.

Research on intelligent fault diagnosis for railway point machines using deep reinforcement learning

Abstract The advanced diagnosis of faults in railway point machines is momentous to ensure the efficient and stable operation of the turnout conversion system. Numerous mature deep learning methods have been extensively applied in this domain. While robust perception has yielded excellent diagnostic outcomes, the deficiency in decision-making capability has led to a lack of overall intelligence. To deal with this, this study proposes an end-to-end deep reinforcement learning (DRL) framework for diagnosing faults in railway point machines. Firstly, a one-dimensional convolutional neural network (1DCNN) is used for the automatic extraction of features from the current signal. Subsequently, the deep Q network (DQN) algorithm is introduced as the core of the diagnostic framework. This involves designing an interactive environment for fault classification and optimizing the agent training network. Finally, leveraging fault data, the agent and the environment engage in continuous interactive learning to produce the ideal classification policy. Multiple comparative experiments are conducted to validate the proposed method. The results demonstrate that the diagnostic accuracy reaches 98.41%, and the average accuracy after many iterations is as high as 99.12%. Notably, this research introduces a creative application of DRL to address the challenge of diagnosing faults in railway point machines. The incorporation of decision thought effectively enhances the intelligence of fault diagnosis.

Structural health monitoring on offshore jacket platforms using a novel ensemble deep learning model

Monitoring health condition of offshore jacket platforms is crucial to prevent unexpected structural damages, where a prevailing challenge involves translating available feature information into structural damage patterns. Although the artificial neural network (ANN) models are popular in addressing this challenge, they often fail to capture the temporal correlations between the feature information and the damage patterns, which reduce their capability for discovering the laws governing the structural damage detection. To bridge this research gap, this study proposes a novel ensemble deep learning model to enhance the temporal feature extraction to improve the damage pattern identification. In this approach, a one-dimensional Convolutional Neural Network (CNN) extracts the spatiotemporal features from the structural vibration measurements. Simultaneously, a SENet attention mechanism is introduced to select the most informatic features. Subsequently, a bidirectional long short-term memory network (BiLSTM) is employed to learn the mapping between the extracted features and the structural damage patterns. Furthermore, the particle swarm optimization (PSO) algorithm is used to optimize the BiLSTM hyperparameters to enhance its stability and reliability. Both simulations and experiments are carried out to collect the vibration responses of the offshore jacket structure in different damage scenarios. The analysis results demonstrate that the proposed method produces remarkable improvement with respect to the accuracy and robustness in identifying the structural damages when compared with the ANNs. The overall detection accuracy of the proposed CNN-BiLSTM-Attention ensemble model is beyond 95%, which provides strong applicability to practical structural health monitoring of offshore platforms.

Vision transformer-based electronic nose for enhanced mixed gases classification

The classification of mixed gases is one of the major functions of the electronic nose. To address the challenges associated with complex feature construction and inadequate feature extraction in gas classification, we propose a classification model for gas mixtures based on the vision transformer (ViT). The whole-process signals of the sensor array are taken as input signals in the proposed classification model, and self-attention mechanism is employed for the fusion of global information and adaptive feature extraction to make full use of the dependence of responses at different stages of the whole-process signals to improve the model’s classification accuracy. Our model exhibited a remarkable accuracy (96.66%) using a dataset containing acetone, methanol, ammonia, and their binary mixtures. In comparison, experiments conducted by support vector machine and a one-dimensional deep convolutional neural network model demonstrated classification accuracy of 90.56% and 92.75%, respectively. Experimental results indicate that the ViT gas classification model can be effectively combined with multi-channel time series data from the sensor array using the self-attention mechanism, thereby improving the accuracy of mixed gases classification. This advancement can be expected to become a standard method for classifying mixed gases.

Multi-degree-of-freedom unmanned aerial vehicle control combining a hybrid brain-computer interface and visual obstacle avoidance

ObjectiveThe difficulty of unmanned aerial vehicle (UAV) control recently lies in multidirectional movement in 3-dimensional space, improving control accuracy and manipulation safety. To address these challenges, a UAV control system that incorporates a hybrid brain-computer interface (hBCI), gyroscope and visual obstacle avoidance based on monocular depth estimation is proposed. Approach. We propose an efficient steady-state visual evoked potential (SSVEP) classification network (CL-NET) featuring a one-dimensional convolutional neural network, a long short-term memory module and an attention module to identify the user's intention for UAV movement in the front, back, left and right directions. The take-off, landing and rising control of the UAV is realized by an electrooculogram (EOG) signal detection algorithm, a blink state detector. In addition, the UAV can fly in an oblique state and rotate according to the current head posture detected by a gyroscope. Furthermore, an improved monocular depth estimation network is employed to design the autonomous obstacle avoidance module of the UAV, ensuring the safety of the brain-controlled system in practice. Main results. The proposed CL-NET delivers an accuracy of 98.67% on the public dataset and an accuracy of 97.92% on the self-collected dataset, both of which surpass the performance of state-of-the-art models. Additionally, we set up a brain control group and a remote control group to conduct practical experiments in a realistic environment. In the experiments involving sixteen subjects, the proposed UAV control system reached an average information transfer rate (ITR) of 44.09 bits/min, and the brain control group had a lower collision rate than the remote control group. Significance. The hybrid control method ensures that the multi-degree-of-freedom (multi-DOF) UAV control system maintains outstanding performance while ensuring good safety.

EEG stress classification based on Doppler spectral features for ensemble 1D-CNN with LCL activation function

The paper proposes an induced stress classification algorithm that uses features from the Doppler spectrum. In this approach, a reference signal source is used to obtain the quadrature and in-phase components of the EEG signal. The higher frequency components from the in-phase and quadrature are then eliminated using a pair of low-pass filters. The Doppler spectrum was then constructed from which the Doppler frequency features are then estimated. The features that were thus obtained are trained using an ensemble 1D-CNN (one-dimensional Convolutional neural network) which uses two sections of 1D-CNN. The first section 1D-CNN trains the features based on the EEG signal classes namely Stroop test, arithmetic task, and mirror tasks, while the second 1D-CNN section trains the features based on the EEG signal intense classes namely high, low, and medium stress. We also propose a linear-cosine-linear (LCL) activation function for the ensemble 1D-CNN which was derived from the cosine signal. The proposed stress classification scheme was evaluated using the SAM-40 datasets with induced stress classes namely arithmetic task, Stroop color-word test, and mirror image recognition task with stress levels namely high, low, and medium with the evaluation metrics such as precision, F1-score, accuracy, specificity, and recall. The proposed stress classification scheme attains an average accuracy, precision, and recall of 95.25%, 95.22%, and 92.9% when evaluated in 9 classes of EEGs.

Toxicity prediction and classification of Gunqile-7 with small sample based on transfer learning method

Drug-induced diseases are the most important component of iatrogenic disease. It is the duty of doctors to provide a reasonable and safe dose of medication. Gunqile-7 is a Mongolian medicine with analgesic and anti-inflammatory effects. As a foreign substance in the body, even with reasonable medication, it may produce varying degrees of adverse reactions or toxic side effects. Since the cost of collecting Gunqile-7 for pharmacological animal trials is high and the data sample is small, this paper employs transfer learning and data augmentation methods to study the toxicity of Gunqile-7. More specifically, to reduce the necessary number of training samples, the data augmentation approach is employed to extend the data set. Then, the transfer learning method and one-dimensional convolutional neural network are utilized to train the network. In addition, we use the support vector machine-recursive feature elimination method for feature selection to reduce features that have adverse effects on model predictions. Furthermore, due to the important role of the pre-trained model of transfer learning, we select a quantitative toxicity prediction model as the pre-trained model, which is consistent with the purpose of this paper. Lastly, the experimental results demonstrate the efficiency of the proposed method. Our method can improve accuracy by up to 9 percentage points compared to the method without transfer learning on a small sample set.

Fault diagnosis and identification of rotating machinery based on one-dimensional convolutional neural network

The paper focuses on two kinds of rotating machinery, miniature table drilling machine and automobile engine, as the research object. Traditional machine learning has the need for manual feature extraction, and is very dependent on expert diagnostic experience and expertise, but also has the disadvantages of low accuracy, low timeliness, low efficiency, etc. For the traditional rotating machinery fault diagnosis method is more based on the traditional machine learning model, this paper puts forward a one-dimensional convolutional neural network-based fault identification method. According to the characteristics of the miniature table drilling machine and the automobile engine which are not detachable, the corresponding faults are set up respectively, Vibration signals of the attitude sensor are obtained by using the signal collector, and the collected data are preprocessed, then the CNN model is built for fault identification, and the network structure is constantly optimized to obtain the optimal network model with high accuracy (up to 100 %) and robustness. The results show that the one-dimensional convolutional neural network model improves the fault recognition accuracy and reduces the cost compared with the traditional machine learning SVM model when the original signal is used as the input signal.

A novel enhanced lemurs optimization-based hybrid attention network for classifying arrthymia from ECG signal using optimal weighted feature

ABSTRACT An arrhythmia is just a disease associated with irregular heart rhythm. Any kind of arrhythmia can be determined with an ECG test. However, this requires long and excellent-quality of ECG samples. This makes an affordable, reliable, and automated system for recognizing and diagnosing arrhythmias. To meet the aforementioned criteria, a new technique with the aid of an attention-aided deep learning approach is proposed to classify arrhythmia in this paper. “Higher-order statistical (HOS)” features are extracted from the original EEG signal and the deep features are obtained from the analyzed QRS complex signals and RR interval by means of a “One-Dimensional Convolution Neural Network (1DCNN)”. The Risk Rate Enhanced Lemurs Optimization Algorithm (RRE-LOA) is utilized to modify the weight, and the extracted features are selected optimally before carrying out the feature selection followed by feature concatenation with weight fusion in order to generate the Weighted Fused Features (WFF). A range of indicators is used to compare the performance of the executed classification model with the traditional models. The results demonstrate that the proposed methodology for classifying arrhythmias offers an enhanced performance while classifying the arrhythmia thus aiding in faster treatment of the arrhythmia disease.

Spectrometer as a quantitative sensor for predicting the weld depth in laser welding

Spectrometers have been used for quality assessment or as auxiliary sensors in multisensor systems to record emissions at different wavelengths. This study developed a single-spectrometer sensor to quantitatively predict the laser penetration depth in Cu/Al overlap and dual-phase (DP) steel bead-on-plate welding joints. The spectrometer signals were sampled at 100 Hz, and an optical coherence tomography (OCT) sensor was used to estimate the in situ penetration depth. A one-dimensional (1D) convolution neural network (CNN) model was developed using the current step of the spectrometer signal. Furthermore, a two-dimensional (2D) CNN model was developed using 10 recent steps of the spectrometer signals. Consequently, 1732 and 3000 data points for DP steel and Al/Cu overlap welding were used for the penetration-depth modeling. Through model training, mean absolute errors of 22.67 and 52.96 µm were achieved for the Al/Cu overlap and DP steel welding, respectively, corresponding to approximately 1.5 % of the substrate thickness. The 2D-CNN models outperformed the 1D-CNN models for both material combinations. However, a slight delay was observed in the response of the models when confronted with abrupt output changes. This study confirmed that spectroscopy can serve as the sole quantitative sensing method for laser welding monitoring and control.

Advancing Non-Cuff Hypertension Detection: Leveraging 1D Convolutional Neural Network and Time Domain Physiological Signals

Timely identification of hypertension (HT) is crucial for effectively managing and reducing the potential health consequences, including cardiovascular events such as heart attacks and strokes, as well as the development of kidney disease. Traditional cuff-based devices often discourage regular monitoring because they cause discomfort. Furthermore, the lack of symptoms in HT complicates the early detection of this condition. To address these challenges, our study employs a non-cuff methodology that utilizes unprocessed electrocardiogram (ECG) and photoplethysmogram (PPG) signals. We utilize a customized approach to enhance the features of a one-dimensional convolutional neural network (CNN) specifically tailored to optimize timeseries data. In contrast to previous research, our methodology avoids the need for complex signal extraction or transformation techniques. The main goal is to identify the optimal input signals and fine-tune the critical hyperparameters of CNNs. The clinical data underwent analysis, which revealed that the use of an integrated ECG and PPG approach resulted in the highest level of accuracy for detection. Notably, the F1 score achieved an impressive value of 98.88%. When evaluated separately, ECG outperformed PPG. Our study contributes to the advancement of the field by introducing a new approach that combines comfort and high accuracy in the early detection of HT. This method is practical and ensures a patient-friendly experience.

Suppressing HIFU interference in ultrasound images using 1D U-Net-based neural networks

Objective. One big challenge with high-intensity focused ultrasound (HIFU) is that the intense acoustic interference generated by HIFU irradiation overwhelms the B-mode monitoring images, compromising monitoring effectiveness. This study aims to overcome this problem using a one-dimensional (1D) deep convolutional neural network. Approach. U-Net-based networks have been proven to be effective in image reconstruction and denoising, and the two-dimensional (2D) U-Net has already been investigated for suppressing HIFU interference in ultrasound monitoring images. In this study, we propose that the one-dimensional (1D) convolution in U-Net-based networks is more suitable for removing HIFU artifacts and can better recover the contaminated B-mode images compared to 2D convolution. Ex vivo and in vivo HIFU experiments were performed on a clinically equivalent ultrasound-guided HIFU platform to collect image data, and the 1D convolution in U-Net, Attention U-Net, U-Net++, and FUS-Net was applied to verify our proposal. Main results. All 1D U-Net-based networks were more effective in suppressing HIFU interference than their 2D counterparts, with over 30% improvement in terms of structural similarity (SSIM) to the uncontaminated B-mode images. Additionally, 1D U-Nets trained using ex vivo datasets demonstrated better generalization performance in in vivo experiments. Significance. These findings indicate that the utilization of 1D convolution in U-Net-based networks offers great potential in addressing the challenges of monitoring in ultrasound-guided HIFU systems.

Deep Learning for Gas Sensing via Infrared Spectroscopy.

Deep learning methods, a powerful form of artificial intelligence, have been applied in a number of spectroscopy and gas sensing applications. However, the speciation of multi-component gas mixtures from infrared (IR) absorption spectra using deep learning remains to be explored. Here, we propose a one-dimensional deep convolutional neural network gas classification model for the identification of small molecules of interest based on IR absorption spectra in flexible user-defined frequency ranges. The molecules considered include ten that are of interest in the atmosphere or in industrial and environmental processes: water vapor, carbon dioxide, ozone, nitrous oxide, carbon monoxide, methane, nitric oxide, sulfur dioxide, nitrogen dioxide, and ammonia. A simulated dataset of IR absorption spectra for mixtures of these molecules diluted in air was generated and used to train a deep learning model. The model was tested against simulated spectra containing noise and was found to provide speciation predictions with accuracy from 82 to 97%. The internal operation of the model was investigated using class activation maps that illustrate how the model prioritizes spectral information for classification. Finally, the model was demonstrated for the prediction of speciation for two synthetic experimental mixture spectra. The proposed model and the dataset generation strategies are generalized and can be implemented for other gases, different frequency ranges, and spectroscopy types. The multi-component speciation method developed herein is the first application of a convolutional neural network model, trained on HITRAN-based simulations, for spectral identification.

A novel concatenated 1D-CNN model for speech emotion recognition

Speech Emotion Recognition (SER) has conventionally been executed solely through acoustic data. Convolutional Neural Network (CNN) in Deep Learning (DL) is a cutting-edge technique that has been effectively used in diverse fields, such as analyzing speech emotion. This study utilizes the Mel Frequency Magnitude Coefficient (MFMC), a modified variant of the Mel Frequency Cepstral Coefficient (MFCC) feature, in conjunction with a one-dimensional (1D) CNN to increase the SER performance. The accuracy of SER is evaluated utilizing the proposed models. Model 1 refers to MFCC-1D-CNN, model 2 refers to MFMC-1D-CNN, and model 3 refers to the concatenated model of model 1 and model 2. The models were evaluated on four datasets comprising eight distinct emotions, specifically anger, happiness, sadness, boredom, fear, surprise, disgust, neutral, and calm. The proposed models, namely model 1, model 2, and model 3, were evaluated on the four datasets using coefficients of 12, 24, and 30. The Proposed Model 1 achieved accuracy rates for EMO-DB, EMOVO, SAVEE, and RAVDESS of 88.1%, 87.5%, 75%, and 97.7% for the respective coefficients. While the proposed model 2 achieved accuracy rates of 93.8%, 98.3%, 90.3%, and 97.5% for the same coefficients. However, the proposed concatenated model 3 outperformed the accuracy of 95.6%, 99.4%, 91.7%, and 98.1% for the four datasets EMO-DB, EMOVO, SAVEE, and RAVDESS compared to model 1 and model 2. Experimental results indicate that the proposed concatenated model 3 improves accuracy and prevents the model from overfitting.

Multimodal 1D CNN for delamination prediction in CFRP drilling process with industrial robots

There is a growing demand for carbon fiber-reinforced plastics (CFRPs) in the aerospace and automotive industries. Consequently, the assembly and repair of CFRP components has garnered considerable attention. However, at industrial sites, there are potential difficulties in the CFRP drilling process. These challenges arise from the anisotropic properties of CFRP and the limitations of using an optical microscope to assess the quality of millions of drilled holes. Therefore, this study introduced an advanced indirect prediction method for the CFRP hole quality based on multisensor data. During the drilling process, data including force, torque, acceleration, voltage, current, sound, and images of the hole exits were acquired via a robotic machining system. The delamination factors, Fd and Fa, which quantify the quality of the hole, can be calculated from the hole images. Preprocessing was employed to segment the sensor data into discrete drilling trials and extract spectral features from the data. This paper proposed a multimodal one-dimensional convolutional neural network (1D CNN) that predicts delamination factors from time series multi-sensor data. A case study, using a test set of 100 trials, validated the proposed model. The performance of the proposed model was evaluated by the mean squared error (MSE) and inference time, yielding 8.42 ± 0.167 × 10−2 and 3.48 ± 0.192 s for Fd, and 6.1 ± 0.729 × 10−2 and 3.13 ± 0.098 s for Fa, respectively, across the entire test set with 100 trials. This result exceeds those of previous studies in terms of both the MSE value and inference time for multimodality, emphasizing the predictive accuracy and real-time operational capabilities of the proposed model in industrial settings. Furthermore, this approach provides practicality and flexibility, facilitating the easy integration or removal of sensors through data-driven preprocessing and a multimodal learning scheme.

Data-driven multivariate regression-based anomaly detection and recovery of unmanned aerial vehicle flight data

Abstract Flight data anomaly detection is crucial to ensuring the safe operation of unmanned aerial vehicles (UAVs) and has been extensively studied. However, the accurate modeling and analysis of flight data is challenging due to the influence of random noise. Meanwhile, existing methods are often inadequate in parameter selection and feature extraction when dealing with large-scale and high-dimensional flight data. This paper proposes a data-driven multivariate regression-based framework considering spatio-temporal correlation for UAV flight data anomaly detection and recovery, which integrates the techniques of correlation analysis (CA), one-dimensional convolutional neural network and long short-term memory (1D CNN-LSTM), and error filtering (EF), named CA-1DCL-EF. Specifically, correlation analysis is first performed on original UAV flight data to select parameters with correlation to reduce the model input and avoid the negative impact of irrelevant parameters on the model. Next, a regression model based on 1D CNN-LSTM is designed to fully extract the spatio-temporal features of UAV flight data and realize parameter mapping. Then, to overcome the effect of random noise, a filtering technique is introduced to smooth the errors to improve the anomaly detection performance. Finally, two common anomaly types are injected into real UAV flight datasets to verify the effectiveness of the proposed method.