Simultaneous localization and mapping algorithm based on point and line features for medical service robots / 第二军医大学学报

Objective To propose a point and line (PL)-simultaneous localization and mapping (SLAM) algorithm and to compare it with oriented FAST and rotated BRIEF (ORB)-SLAM2, so as to improve the global localization accuracy and real-time performance of SLAM algorithm for medical service robots. Methods The PL-SLAM algorithm added line features based on point feature in the process of feature extraction, and carried out mapping and global localization in the complex medical environment according to the point and line features after fusion. The public datasets (EuRoc and KITTI) were used to compare the PL-SLAM and ORB-SLAM2 algorithms, and the comprehensive performance of autonomous navigation of the medical service robots was tested. Results Compared with the ORB-SLAM2 algorithm, PL-SLAM algorithm extracted more point and line features in weak texture scenario, and effectively enhanced the global localization accuracy and real-time performance. The rotation error of the PL-SLAM algorithm decreased by 42.2% and the runtime increased by 55.9%. Conclusion PL-SLAM algorithm can effectively improve global localization accuracy and the real-time performance of medical service robots.

Application of least square support vector machine algorithm in clinical pulse diagram parameter-blood pressure prediction model of traditional Chinese medicine / 第二军医大学学报

Objective To propose a learning model based on least square support vector machine (LSSVM) algorithm to improve the accuracy and efficiency for predicting clinical blood pressure data of traditional Chinese medicine (TCM). Methods The LSSVM learning model was used to predict the clinical blood pressure of TCM. By replacing the inequality constraints of support vector machine with LSSVM equality constraints, the quadratic programming problem was transformed into a linear equation solution problem to reduce computational complexity and speed up algorithm convergence. The clinical pulse diagram parameters and blood pressure data of 320 patients were collected. Three hundred of them were used as training samples, the remaining 20 samples were used as test data. The LSSVM learning model was used to predict blood pressure data according to the pulse diagram parameters of the patients. Results Experimental results showed that the LSSVM learning model had high prediction accuracy for blood pressure data. The LSSVM learning model based on polynomial kernel function had better learning and prediction abilities than the LSSVM learning model based on radial basis kernel function. The mean prediction errors of systolic blood pressure, diastolic blood pressure and mean arterial pressure obtained by the LSSVM learning model based on polynomial kernel function were 7.88%, 8.40% and 6.67%, respectively, which were lower than those obtained by the LSSVM learning model based on radial basis kernel function (7.95%, 9.70% and 7.48%, respectively). Conclusion The LSSVM learning model proposed in this experiment can be used to predict the blood pressure data of patients only by the clinical pulse diagram parameters, and is a good reference for clinical diagnosis of TCM.

Tongue image classification method based on transfer learning and fully connected neural network / 第二军医大学学报

Objective To propose a classification method for small sample tongue images based on transfer learning and fully connected neural network, so as to solve the problems of large amount of data, high requirement of training equipment and long training time of deep learning in the classification of tongue images. Methods Effective features such as tongue points and lines of tongue images were extracted by the convolution Inception_v3 network after training on the massive data set of ImageNet. The above features were classified by the fully connected neural network, and the image knowledge acquired by the deep learning network was transferred to the tongue image recognition task, and then the tongue data set were used to train and test the efficiency of the network. Results Compared with the typical tongue image classification method such as K-nearest neighbor (KNN) algorithm, support vector machine (SVM) algorithm and convolutional neural network (CNN) deep learning method, the two methods (Inception_v3＋2NN and Inception_v3＋3NN) in our experiment had higher classification rates for tongue images, with the accuracy rates being 90.30% and 93.98%, respectively, and had shorter training time for the sample. Conclusion Compared with KNN algorithm, SVM algorithm and CNN deep learning method, the tongue image classification method based on transfer learning and fully connected neural network can effectively improve the accuracy rate of tongue image classification and shorten the training time.

Application of feature matching algorithm based on grid-based motion statistics in medical service robot / 第二军医大学学报

Objective To propose a scoring framework grid-based motion statistics (SF-GMS) feature matching algorithm to improve the poor real-time ability and inaccurate matching in the process of target recognition for medical service robots. Methods The feature point neighborhoods were segmented by SF-GMS algorithm using the grids, and the number of feature points in each neighborhood was counted and the scoring frame function was set to judge the feature matching accuracy according to the number of neighborhood feature points and the scoring threshold. Results and conclusion Compared with random sample consensus algorithm, SF-GMS algorithm effectively improved the successful matching rate, and had better real-time performance. SF-GMS algorithm had better stability to the changes of illumination view, occlusion, affine, scale and rotation, and could meet the demand of autonomous navigation in simulating hospital ward scenario for medical service robots.