Training Of Convolutional Neural Network Model Research Articles

Classification of nutmeg ripeness using artificial intelligence

Nutmeg seeds can produce a lot of oil if they have optimal maturity, in other words, they have little moisture content. Based on observations I made at one of the refineries in Sukabumi, farmers do not pay attention to the maturity level of nutmeg seeds after drying which can cause a decrease in the quality of nutmeg seeds and the quality of the oil produced. This study aims to make it easier for nutmeg farmers to classify the maturity of nutmeg seeds. This study used the Convolutional Neural Network (CNN) method to help with classification problems and several image processing methods. This program will be run through an android application. The results of CNN model training accuracy are 97.92%. Thus, it can be concluded that the design and testing of a model to classify the maturity level of nutmeg seeds using artificial intelligence and the implementation of the model into an android application has been successfully carried out.

Self-avatar: Monocular 3D human reconstruction from RGB image

3D human position and shape estimate are crucial in many computer vision applications. Despite the fact that there are numerous deep learning techniques designed to handle this problem, they frequently only use training networks with RGB images from a single point of view. In this paper, a unique approach to solve this issue is proposed by combining a regression-based multi-view picture learning loop with an optimization-based multi-view model. This is because some public datasets are collected by multi-view camera systems. A parameterized human body model's position and shape parameters are initially deduced by a convolutional neural network (CNN) from multi-view photos. This work then introduces an enhanced multi-view optimization method called MV-SMPLify, which aligns the SMPL model with multi-view images by using the regressed pose and shape as beginning values. Following that, the CNN model's training can be monitored using the optimum parameters. The Self-avatar project as a whole is a self-supervised framework that combines the advantages of both the CNN method and the optimization-based strategy. Additionally, the use of multi-view photos improves thorough supervision during training. This methodology outperforms earlier methods in a variety of ways, according to qualitative and quantitative testing using open datasets.

Deteksi Penyakit Pada Daun Tanaman Ubi Jalar Menggunakan Metode Convolutional Neural Network

Sweet potatoes are the world's third most important root crop and the fourth most popular staple food in developing countries, including Indonesia. Some diseases commonly found in sweet potato leaves are early blight (identified by leaves containing batataezim) and late blight (characterized by leaves that have chlorosis). These two diseases have different symptoms and require different treatments, but a slow identification process can lead to additional costs for plant care. This research will classify image data of sweet potato diseases using the Convolutional Neural Network (CNN) method. CNN is a derivative of the Multilayer Perceptron (MLP) designed to process image data with high network depth and is often used for classification tasks. The research uses a total of 750 images divided into 3 classes: images of healthy leaves, images of leaves with chlorosis, and images of leaves containing batataezim. Each leaf class will be labeled with 250 image data, and the labeled data will be further divided into training and testing sets. From these sets, prediction data will be obtained from the testing process during the CNN model training. The training accuracy resulted in a value of 98.17%, while the testing accuracy reached 98.67%. Additionally, the resulting loss values are remarkably low, at 0.04% for training and 0.03% for testing. The research findings will provide insights into the CNN method's ability to detect diseases in sweet potato plants, potentially impacting agricultural supervision, plant disease identification, and enabling more precise decisions regarding plant care actions.

Wearable edge machine learning with synthetic photoplethysmograms

Strict privacy regulations pose challenges to the development of machine learning (ML) in the field of health technology where data is particularly sensitive. Gathering and using robust, bias-free, and suitably anonymized datasets required by ML models is difficult, time-consuming, and thus expensive. Parametric synthetic data offers a solution by mimicking real-world processes with easily adjustable parameters that shape the information content of the data as desired. This article presents a system demonstrating how synthetic data can be used in conjunction with wearable edge devices. Importantly, the system preserves privacy as there is no risk of leaking sensitive information from the model or during the use of the wearable device. The system consists of (1) a synthetic photoplethysmogram (PPG) model, (2) convolutional neural network (CNN) models trained with the synthetic signals, (3) a wearable edge device that computes heart rate from real-time PPG signals using the developed CNN models, and (4) an accompanying mobile phone application receiving the results. The synthetic model produces realistic PPG signals together with labels that can be used in CNN model training. The quality of the synthetic data is sufficient to train even a tiny CNN model with only two convolutional layers and 28 parameters to detect PPG waveform feet. The developed wearable device is able to run the model smoothly and the performance of the model is on par with the more complex models and other foot detection algorithms.

Bimodal CNN for cardiovascular disease classification by co-training ECG grayscale images and scalograms

This study aimed to develop a bimodal convolutional neural network (CNN) by co-training grayscale images and scalograms of ECG for cardiovascular disease classification. The bimodal CNN model was developed using a 12-lead ECG database collected from Chapman University and Shaoxing People's Hospital. The preprocessed database contains 10,588 ECG data and 11 heart rhythms labeled by a specialist physician. The preprocessed one-dimensional ECG signals were converted into two-dimensional grayscale images and scalograms, which are fed simultaneously to the bimodal CNN model as dual input images. The proposed model aims to improve the performance of CVDs classification by making use of ECG grayscale images and scalograms. The bimodal CNN model consists of two identical Inception-v3 backbone models, which were pre-trained on the ImageNet database. The proposed model was fine-tuned with 6780 dual-input images, validated with 1694 dual-input images, and tested on 2114 dual-input images. The bimodal CNN model using two identical Inception-v3 backbones achieved best AUC (0.992), accuracy (95.08%), sensitivity (0.942), precision (0.946) and F1-score (0.944) in lead II. Ensemble model of all leads obtained AUC (0.994), accuracy (95.74%), sensitivity (0.950), precision (0.953), and F1-score (0.952). The bimodal CNN model showed better diagnostic performance than logistic regression, XGBoost, LSTM, single CNN model training with grayscale images alone or with scalograms alone. The proposed bimodal CNN model would be of great help in diagnosing cardiovascular diseases.

Quantitative Mass Spectrometry Imaging Using Multivariate Curve Resolution and Deep Learning: A Case Study

In the present contribution, a novel approach based on multivariate curve resolution and deep learning (DL) is proposed for quantitative mass spectrometry imaging (MSI) as a potent technique for identifying different compounds and creating their distribution maps in biological tissues without need for sample preparation. As a case study, chlordecone as a carcinogenic pesticide was quantitatively determined in mouse liver using matrix-assisted laser desorption ionization-MSI (MALDI-MSI). For this purpose, data from seven standard spots containing 0 to 20 picomoles of chlordecone and four unknown tissues from the mouse livers infected with chlordecone for 1, 5, and 10 days were analyzed using a convolutional neural network (CNN). To solve the lack of sufficient data for CNN model training, each pixel was considered as a sample, the designed CNN models were trained by pixels in training sets, and their corresponding amounts of chlordecone were obtained by multivariate curve resolution-alternating least-squares (MCR-ALS). The trained models were then externally evaluated using calibration pixels in test sets for 1, 5, and 10 days of exposure, respectively. Prediction R2 for all three data sets ranged from 0.93 to 0.96, which was superior to support vector machine (SVM) and partial least-squares (PLS). The trained CNN models were finally used to predict the amount of chlordecone in mouse liver tissues, and their results were compared with MALDI-MSI and GC-MS methods, which were comparable. Inspection of the results confirmed the validity of the proposed method.

Sound Recognition based on Convolutional Neural Network for Real-Time Cutting State Monitoring of Tube Cutting Machine

Machine sound monitoring is widely used in various applications of operational state and diagnostic monitoring as machine-emitted sound contains the operational and process information. In the metal cutting industry, it is not surprising that operators are easily able to recognize whether cutting is engaging by listening to the operational sounds based on their experiences even if the cutting parameters are changed. Inspired by the ability of recognizing human sound, we propose a real-time sound monitoring for cutting state based on the convolutional neural network (CNN). Three sound sensors were deployed to a tube cutting machine on different places to capture and analyze sound signals in real-time. To enable sound stream automatically, the MTConnect framework was employed. Four different sizes: 0.25, 0.5, 0.75, and 1-second, of the time frames, were selected and the effects of the time frames, as well as the sensors, were compared in terms of performances of predicting the cutting time and the productivity. The Log-Mel spectrum was adopted as a feature for convolutional neural network (CNN) model. In the CNN model training, a random search method was used to determine the hyperparameters of the CNN models. To verify the trained CNN models, the evaluation dataset which consisted of unknown and untrained part production was used. The accuracies of predicted cutting state on the evaluation dataset for all sensors and all sizes of the time frames ranged from 83.3 to 98.6%. The accuracies of the predicted productivity were also analyzed according to the sensors and the size of time frames. Furthermore, real-time monitoring response was evaluated to figure out the effects of the size of time frames on the prediction speed and the time delay. The sizes of the time frames play an important role in various aspects. As a result, it is suggested that the shorter size of the time frames is the better choice if the prediction accuracy of cutting time and the monitoring response time is critical whereas the longer size is the better if the prediction accuracy of productivity is more important. Among the internal sound sensors in various locations, one on the machine base shows the best performances in all sizes of the time frames with respect to not only the prediction of the cutting time but also the performance of predicting the productivity.

Application of Fault Overlay Method and CNN in Infrared Image of Detecting Inter-Turn Short-Circuit in Dry-Type Transformer

Inter-turn short-circuit (ITSC) faults do not necessarily produce high temperatures but have special heat distribution and characteristics. Therefore, a new recognition solution for diagnosing faults is proposed, based on the fault overlay method, and the convolutional neural network (CNN) is trained to achieve the automatic identification of infrared images. In this method, through the coverage of layers, the proposed image augmentation method is carried out and simulates the fault data of increasing training. We produce 43 fault traces through the fault overlay method on the three-phase winding of a transformer and use 90 infrared images of transformers in normal operation combined with them to enhance the amount of image data. The fault recognition ability is realized based on CNN model training, including analysis of experimental results of grayscale and color images, and Gaussian noise. In the test of the practical case, a short-circuit test of the 11.4 kV dry-type transformer is carried out, and the ITSC fault is identified when the load is about 15%. The fault characteristic block on this thermal image is 36.3 degrees, which verifies the identification available by this method, and has a certain reference value for the development of infrared image diagnosis technology for power equipment.

Intelligent identification of film on cotton based on hyperspectral imaging and convolutional neural network.

The identification of the film on cotton is of great significance for the improvement of cotton quality. Most of the existing technologies are dedicated to removing colored foreign fibers from cotton using photoelectric sorting methods. However, the current technologies are difficult to identify colorless transparent film, which becomes an obstacle for the harvest of high-quality cotton. In this paper, an intelligent identification method is proposed to identify the colorless and transparent film on cotton, based on short-wave near-infrared hyperspectral imaging and convolutional neural network (CNN). The algorithm includes black-and-white correction of hyperspectral images, hyperspectral data dimensionality reduction, CNN model training and testing. The key technology is that the features of the hyperspectral image data are degraded by the principal component analysis (PCA) to reduce the amount of computing time. The main innovation is that the colorless and transparent film on cotton can be accurately identified through a CNN with the performance of automatic feature extraction. The experimental results show that the proposed method can greatly improve the identification precision, compared with the traditional methods. After the simulation experiment, the method proposed in this paper has a recognition rate of 98.5% for film. After field testing, the selection rate of film is as high as 96.5%, which meets the actual production needs.

PPCNN: An efficient privacy‐preserving CNN training and inference framework

Convolutional neural network (CNN) is one of the representative models of deep learning, commonly used to analyze visual images. CNN model is more accurate when trained on large amounts of data from multiple sources, and the huge training cost makes the model much more valuable. However, data from various sources is often privacy-sensitive. Therefore, the privacy of these data should be protected during CNN model training and inference. In this paper, we propose an efficient and secure two-party computation (2PC) framework PPCNN for privacy-preserving CNN training and inference. Specifically, we use a new secret sharing technique introduced in ABY2.0 to securely compute various computational tasks involved in the CNN training and inference processes. This secret sharing technique can significantly reduce the communication overhead. Meanwhile, we assign these computationally intensive tasks to cloud servers to reduce the computational burden on local devices. We demonstrate the security of these protocols in the semihonest model. In addition, we use the MP-SPDZ library to simulate our PPCNN framework, and the experiments prove its high efficiency and accuracy.

Masked Face Recognition and Temperature Monitoring Systems for Airplane Passenger Using Sensor Fusion

Transportation is currently an unavoidable necessity. However, the COVID-19 pandemic has impacted all lines of industry, including the Indonesian aviation transportation industry. Technology is one of the solutions to deal with these problems. The monitoring system of masked face recognition and body temperature detection for the check-in process of passengers at the airport is aimed to be developed in this research. The contribution of this research is that the system can distinguish the type of face mask used. Therefore, this monitoring system classified only medical masks and N95/KN95 respirator masks as ‘Good Masked’. IP camera and thermal camera are used to identify a masked face and body temperature, respectively. The sensor fusion method was used for decision-making on passengers whether they can be departed or not. The decision was taken based on the measured body temperature, the use of standardized face masks, and the face recognition of the airport passengers. Convolutional neural network (CNN) method was used for face and face mask recognition. The CNN model training was conducted four times according to the four proposed scenarios. The CNN model that has been trained can distinguish a masked face and a face without a mask. The best results were obtained in the fourth scenario with the comparison of the training dataset to the testing dataset was 9:1 and the epoch was 500 times. The basic deep learning model used for face detection was the single shot multibox detector (SSD) using the ResNet-10 architecture. Meanwhile, the CNN method with the MobileNetV2 architecture was used to detect the use of masks. The accuracy of the CNN model for face recognition and mask recognition was 100%. All check-in monitoring and verification process data were displayed on the web application which was built on the localhost.

Structure health monitoring of a composite wing based on flight load and strain data using deep learning method

An investigation was made into a method for Structural Health Monitoring (SHM) of a composite wing using Convolutional Neural Network (CNN) model. In this method, various aerodynamic loads of an aircraft during flight and corresponding strain data were used for CNN model training. The proposed method was demonstrated by numerical simulation using vortex lattice method for aerodynamic loads of an A350-type aircraft in over a thousand flight conditions and a Finite Element (FE) model as a digital twin of the full-scale composite wing. To represent the measurement of 324 sensors mounted in the 18 skin-rib joints of the inboard wing, strain data from the 18 × 18 elements of the FE model in the sensor locations were calculated corresponding to the flight loadings. The strain data from the original structure FE model were employed to train a CNN model that was classified as healthy samples. Damaged elements were then introduced in random locations to produce data samples corresponding to the same set of flight loads for the CNN model training. In the subsequent damage detection process using the trained CNN model, confusion matrix, uncertainty and sensitivity analysis were evaluated. The study results show that robust damage detection results can be obtained with 99% accuracy without noise and 97% accuracy with 2% Gaussian noise. In the damage localization process, threshold value was set at 1.5, 2 or 2.5, and 83% overall accuracy was achieved using the CNN model when the threshold value was 1.5. The study demonstrated that the proposed method is efficient, accurate and robust.

Deep learning-based automatic delineation of anal cancer gross tumour volume: a multimodality comparison of CT, PET and MRI

Background Accurate target volume delineation is a prerequisite for high-precision radiotherapy. However, manual delineation is resource-demanding and prone to interobserver variation. An automatic delineation approach could potentially save time and increase delineation consistency. In this study, the applicability of deep learning for fully automatic delineation of the gross tumour volume (GTV) in patients with anal squamous cell carcinoma (ASCC) was evaluated for the first time. An extensive comparison of the effects single modality and multimodality combinations of computed tomography (CT), positron emission tomography (PET), and magnetic resonance imaging (MRI) have on automatic delineation quality was conducted. Material and methods 18F-fluorodeoxyglucose PET/CT and contrast-enhanced CT (ceCT) images were collected for 86 patients with ASCC. A subset of 36 patients also underwent a study-specific 3T MRI examination including T2- and diffusion-weighted imaging. The resulting two datasets were analysed separately. A two-dimensional U-Net convolutional neural network (CNN) was trained to delineate the GTV in axial image slices based on single or multimodality image input. Manual GTV delineations constituted the ground truth for CNN model training and evaluation. Models were evaluated using the Dice similarity coefficient (Dice) and surface distance metrics computed from five-fold cross-validation. Results CNN-generated automatic delineations demonstrated good agreement with the ground truth, resulting in mean Dice scores of 0.65–0.76 and 0.74–0.83 for the 86 and 36-patient datasets, respectively. For both datasets, the highest mean Dice scores were obtained using a multimodal combination of PET and ceCT (0.76–0.83). However, models based on single modality ceCT performed comparably well (0.74–0.81). T2W-only models performed acceptably but were somewhat inferior to the PET/ceCT and ceCT-based models. Conclusion CNNs provided high-quality automatic GTV delineations for both single and multimodality image input, indicating that deep learning may prove a versatile tool for target volume delineation in future patients with ASCC.

Deep learning from imbalanced data for automatic defect detection in multicrystalline solar wafer images

This paper presents a deep learning method for automated defect inspection in multicrystalline solar wafer surfaces. A multicrystalline solar wafer contains local crystal grains with random shapes, sizes and gray-values. It shows a heterogeneous texture in the surface, and makes the automated optical inspection task very challenging. The conventional machine vision needs individually handcraft specific texture features for different types of defects in the silicon solar wafer. The deep learning technique is an end-to-end training. It can learn discriminant features and classification simultaneously. However, it needs a large amount of defect-free and defective samples for the model training. The defect samples are generally not sufficient in the manufacturing process. A generative adversarial network based model is used to generate synthetic defect samples from a small set of real defects. The synthesized defective and true defect-free samples are then used to train a convolutional neural network (CNN) model for defect detection. It overcomes the imbalanced class data problem arising in the CNN model training. The proposed method is compared with the commonly-used learning strategies for imbalanced data. Experimental results show that the proposed deep learning approach can be applied to a variety of defect types found in the multicrystalline solar wafer surface.

Quantitative analysis of lithium in brine by laser-induced breakdown spectroscopy based on convolutional neural network

In this study, a simple and effective method for accurate determination of lithium in brine samples was developed by the combination of laser induced breakdown spectroscopy (LIBS) and convolutional neural network (CNN). Our results clearly demonstrate that the use of CNN could efficiently overcome the complex matrix effects, and thus allows for on-site Li quantitative determination in brine samples by LIBS. Specifically, two CNN models with different input data (M-CNN with matrix emission lines, and DP-CNN with double Li lines) were constructed based on the primary matrix features on spectrum and Boltzmann equation, respectively. It was observed that DP-CNN model could greatly improve the accuracy of Li analysis. We also compared the quantitative analysis capabilities of DP-CNN model with partial least squares regression (PLSR) and principal component analysis-support vector regression (PCA-SVR) model, and the results clearly showed DP-CNN offers the best quantification results (higher accuracy and less matrix interference). Finally, five real brine samples were successfully analyzed by the proposed DP-CNN model, confirming by the average absolute error of the prediction of 0.28 mg L−1 and the average relative error of 3.48%. These results clearly demonstrate that input data plays an important role in the training of CNN model in LIBS analysis, and the proposed DP-CNN provides an effective approach to solve the matrix effects encountered in LIBS for Li measurement in brine samples.

Codeless Deep Learning of COVID-19 Chest X-Ray Image Dataset with KNIME Analytics Platform.

ObjectivesThis paper proposes a method for computer-assisted diagnosis of coronavirus disease 2019 (COVID-19) through chest X-ray imaging using a deep learning model without writing a single line of code using the Konstanz Information Miner (KNIME) analytics platform.MethodsWe obtained 155 samples of posteroanterior chest X-ray images from COVID-19 open dataset repositories to develop a classification model using a simple convolutional neural network (CNN). All of the images contained diagnostic information for COVID-19 and other diseases. The model would classify whether a patient was infected with COVID-19 or not. Eighty percent of the images were used for model training, and the rest were used for testing. The graphic user interface-based programming in the KNIME enabled class label annotation, data preprocessing, CNN model training and testing, performance evaluation, and so on.Results1,000 epochs training were performed to test the simple CNN model. The lower and upper bounds of positive predictive value (precision), sensitivity (recall), specificity, and f-measure are 92.3% and 94.4%. Both bounds of the model’s accuracies were equal to 93.5% and 96.6% of the area under the receiver operating characteristic curve for the test set.ConclusionsIn this study, a researcher who does not have basic knowledge of python programming successfully performed deep learning analysis of chest x-ray image dataset using the KNIME independently. The KNIME will reduce the time spent and lower the threshold for deep learning research applied to healthcare.

Predicting the central 10 degrees visual field in glaucoma by applying a deep learning algorithm to optical coherence tomography images

We aimed to develop a model to predict visual field (VF) in the central 10 degrees in patients with glaucoma, by training a convolutional neural network (CNN) with optical coherence tomography (OCT) images and adjusting the values with Humphrey Field Analyzer (HFA) 24–2 test. The training dataset included 558 eyes from 312 glaucoma patients and 90 eyes from 46 normal subjects. The testing dataset included 105 eyes from 72 glaucoma patients. All eyes were analyzed by the HFA 10-2 test and OCT; eyes in the testing dataset were additionally analyzed by the HFA 24-2 test. During CNN model training, the total deviation (TD) values of the HFA 10-2 test point were predicted from the combined OCT-measured macular retinal layers’ thicknesses. Then, the predicted TD values were corrected using the TD values of the innermost four points from the HFA 24-2 test. Mean absolute error derived from the CNN models ranged between 9.4 and 9.5 B. These values reduced to 5.5 dB on average, when the data were corrected using the HFA 24-2 test. In conclusion, HFA 10-2 test results can be predicted with a OCT images using a trained CNN model with adjustment using HFA 24-2 test.

Fault and horizon automatic interpretation by CNN: a case study of coalfield

Abstract A convolutional neural network (CNN) is a powerful tool used for seismic interpretation. It does not require manual intervention and can automatically detect geological structures using the pattern features of the original seismic data. In this study, we presented the development history of seismic interpretation and the application of CNN in seismic exploration. We proposed a set of CNN prediction methods and processes for coalfield seismic interpretation and realised automatic interpretation of faults and horizons based on the relationship between faults and horizons. We defined a CNN model training method based on structural geological modelling, which allowed rapid and accurate establishment of fault and horizon labels by using structural modelling. We used two examples to verify the accuracy of the algorithm, one to test for synthetic 3D seismic data and one to test for real coalfield seismic data. The results showed that CNNs can effectively predict both faults and horizons at the same time and has high accuracy. Thus, CNNs are potentially novel interpretation tools for coalfield seismic interpretation.

Application of artificial intelligence using a novel EUS-based convolutional neural network model to identify and distinguish benign and malignant hepatic masses

Detection and characterization of focal liver lesions (FLLs) is key for optimizing treatment for patients who may have a primary hepatic cancer or metastatic disease to the liver. This is the first study to develop an EUS-based convolutional neural network (CNN) model for the purpose of identifying and classifying FLLs. A prospective EUS database comprising cases of FLLs visualized and sampled via EUS was reviewed. Relevant still images and videos of liver parenchyma and FLLs were extracted. Patient data were then randomly distributed for the purpose of CNN model training and testing. Once a final model was created, occlusion heatmap analysis was performed to assess the ability of the EUS-CNN model to autonomously identify FLLs. The performance of the EUS-CNN for differentiating benign and malignant FLLs was also analyzed. A total of 210,685 unique EUS images from 256 patients were used to train, validate, and test the CNN model. Occlusion heatmap analyses demonstrated that the EUS-CNN model was successful in autonomously locating FLLs in 92.0% of EUS video assets. When evaluating any random still image extracted from videos or physician-captured images, the CNN model was 90% sensitive and 71% specific (area under the receiver operating characteristic [AUROC], 0.861) for classifying malignant FLLs. When evaluating full-length video assets, the EUS-CNN model was 100% sensitive and 80% specific (AUROC, 0.904) for classifying malignant FLLs. This study demonstrated the capability of an EUS-CNN model to autonomously identify FLLs and to accurately classify them as either malignant or benign lesions.

Shoulder muscle activation pattern recognition based on sEMG and machine learning algorithms

Background and ObjectiveSurface electromyography (sEMG) has been used for robotic rehabilitation engineering for volitional control of hand prostheses or elbow exoskeleton, however, using sEMG for volitional control of an upper limb exoskeleton has not been perfectly developed. The long-term goal of our study is to process shoulder muscle bio-electrical signals for rehabilitative robotic assistive device motion control. The purposes of this study included: 1) to test the feasibility of machine learning algorithms in shoulder motion pattern recognition using sEMG signals from shoulder and upper limb muscles, 2) to investigate the influence of motion speed, individual variability, EMG recording device, and the amount of EMG datasets on the shoulder motion pattern recognition accuracy. MethodsA novel convolutional neural network (CNN) structure was constructed to process EMG signals from 12 muscles for the pattern recognition of upper arm motions including resting, drinking, backward-forward motion, and abduction motion. The accuracy of the CNN models for pattern recognition under different motion speeds, among individuals, and by EMG recording devices was statistically analyzed using ANOVA, GLM Univariate analysis, and Chi-square tests. The influence of EMG dataset number used for CNN model training on recognition accuracy was studied by gradually increasing dataset number until the highest accuracy was obtained. ResultsResults showed that the accuracy of the normal speed CNN model in motion pattern recognition was 97.57% for normal speed motions and 97.07% for fast speed motions. The accuracy of the cross-subjects CNN model in motion pattern recognition was 79.64%. The accuracy of the cross-device CNN model in motion pattern recognition was 88.93% for normal speed motion and 80.87% for mixed speed. There was a statistical difference in pattern recognition accuracy between different CNN models. ConclusionThe EMG signals of shoulder and upper arm muscles from the upper limb motions can be processed using CNN algorithms to recognize the identical motions of the upper limb including drinking, forward/backward, abduction, and resting. A simple CNN model trained by EMG datasets of a designated motion speed accurately detected the motion patterns of the same motion speed, yielding the highest accuracy compared with other mixed CNN models for various speeds of motion pattern recognition. Increase of the number of EMG datasets for CNN model training improved the pattern recognition accuracy.