In a telemedicine diagnosis system, the emergence of 3D imaging enables doctors to make clearer judgments, and its accuracy also directly affects doctors’ diagnosis of the disease. In order to ensure the safe transmission and storage of medical data, a 3D medical watermarking algorithm based on wavelet transform is proposed in this paper. The proposed algorithm employs the principal component analysis (PCA) transform to reduce the data dimension, which can minimize the error between the extracted components and the original data in the mean square sense. Especially, this algorithm helps to create a bacterial foraging model based on particle swarm optimization (BF-PSO), by which the optimal wavelet coefficient is found for embedding and is used as the absolute feature of watermark embedding, thereby achieving the optimal balance between embedding capacity and imperceptibility. A series of experimental results from MATLAB software based on the standard MRI brain volume dataset demonstrate that the proposed algorithm has strong robustness and make the 3D model have small deformation after embedding the watermark.

The leakage of medical audio data in telemedicine seriously violates the privacy of patients. In order to avoid the leakage of patient information in telemedicine, a two-stage reversible robust audio watermarking algorithm is proposed to protect medical audio data. The scheme decomposes the medical audio into two independent embedding domains, embeds the robust watermark and the reversible watermark into the two domains respectively. In order to ensure the audio quality, the Hurst exponent is used to find a suitable position for watermark embedding. Due to the independence of the two embedding domains, the embedding of the second-stage reversible watermark will not affect the first-stage watermark, so the robustness of the first-stage watermark can be well maintained. In the second stage, the correlation between the sampling points in the medical audio is used to modify the hidden bits of the histogram to reduce the modification of the medical audio and reduce the distortion caused by reversible embedding. Simulation experiments show that this scheme has strong robustness against signal processing operations such as MP3 compression of 48 db, additive white Gaussian noise (AWGN) of 20 db, low-pass filtering, resampling, re-quantization and other attacks, and has good imperceptibility.

This paper proposes a soft tissue grasping deformation model, where BP neural network optimized by the genetic algorithm is used to realize the real-time and accurate interaction of soft tissue grasping during virtual surgery. In the model, the soft tissue epidermis is divided into meshes, and the meshes generate displacements under the action of tension. The relationship between the tension and displacement of the mesh is determined by the proposed cylindrical spiral spring model. The optimized BP neural network is trained based on the sample data of the mesh point and vertical tension, so as to obtain the force and displacement of any mesh point on the soft tissue epidermis. The virtual experiment platform is built using a PHANTOM OMNI haptic hand controller and the 3D Max software, by which the simulation experiment of grasping the human abdomen is realized. The experimental results show that the proposed model has good visual interaction and real-time force feedback, which can meet the requirements of deformation simulation for soft tissue grasping in virtual surgery.

The key to preventing the COVID-19 is to diagnose patients quickly and accurately. Studies have shown that using Convolutional Neural Networks (CNN) to analyze chest Computed Tomography (CT) images is helpful for timely COVID-19 diagnosis. However, personal privacy issues, public chest CT data sets are relatively few, which has limited CNN's application to COVID-19 diagnosis. Also, many CNNs have complex structures and massive parameters. Even if equipped with the dedicated Graphics Processing Unit (GPU) for acceleration, it still takes a long time, which is not conductive to widespread application. To solve above problems, this paper proposes a lightweight CNN classification model based on transfer learning. Use the lightweight CNN MobileNetV2 as the backbone of the model to solve the shortage of hardware resources and computing power. In order to alleviate the problem of model overfitting caused by insufficient data set, transfer learning is used to train the model. The study first exploits the weight parameters trained on the ImageNet database to initialize the MobileNetV2 network, and then retrain the model based on the CT image data set provided by Kaggle. Experimental results on a computer equipped only with the Central Processing Unit (CPU) show that it consumes only 1.06 s on average to diagnose a chest CT image. Compared to other lightweight models, the proposed model has a higher classification accuracy and reliability while having a lightweight architecture and few parameters, which can be easily applied to computers without GPU acceleration. Code:github.com/ZhouJie-520/paper-codes.

With the increasing application of surveillance cameras, vehicle re-identification (Re-ID) has attracted more attention in the field of public security. Vehicle Re-ID meets challenge attributable to the large intra-class differences caused by different views of vehicles in the traveling process and obvious inter-class similarities caused by similar appearances. Plentiful existing methods focus on local attributes by marking local locations. However, these methods require additional annotations, resulting in complex algorithms and insufferable computation time. To cope with these challenges, this paper proposes a vehicle Re-ID model based on optimized DenseNet121 with joint loss. This model applies the SE block to automatically obtain the importance of each channel feature and assign the corresponding weight to it, then features are transferred to the deep layer by adjusting the corresponding weights, which reduces the transmission of redundant information in the process of feature reuse in DenseNet121. At the same time, the proposed model leverages the complementary expression advantages of middle features of the CNN to enhance the feature expression ability. Additionally, a joint loss with focal loss and triplet loss is proposed in vehicle Re-ID to enhance the model’s ability to discriminate difficult-to-separate samples by enlarging the weight of the difficult-to-separate samples during the training process. Experimental results on the VeRi-776 dataset show that mAP and Rank-1 reach 75.5% and 94.8%, respectively. Besides, Rank-1 on small, medium and large sub-datasets of Vehicle ID dataset reach 81.3%, 78.9%, and 76.5%, respectively, which surpasses most existing vehicle Re-ID methods.

Fall behavior is closely related to the high mortality rate of the elderly, so fall detection has become an important and urgent research area in human behavior recognition. However, the existing fall detection methods, suffer from the loss of detailed action information during feature extraction due to the downsampling operation, resulting in subpar performance when detecting falls with similar behaviors such as lying and sitting. To solve the challenges, this study proposes a high-resolution spatio-temporal feature extraction method based on a spatio-temporal coordinate attention mechanism. The method employs 3D convolutions to extract spatio-temporal features and utilizes gradual down-sampling to generate a multi-resolution sub-network, thus realizing multi-scale fusion and perception enhancement of details. In particular, this study designs a pseudo-3D basic block, which simulates the ability of 3D convolution, to ensure the running speed of the network while controlling the number of parameters. Further, a spatio-temporal coordinate attention mechanism is designed to accurately extract the spatio-temporal positional changes of key skeletal points and the interrelationships among them. Long-term dependencies in horizontal, vertical, temporal directions are captured through three one-dimensional global pooling operations. Then the long-range relationships and channel correlations among features are captured by cascading and slicing operations. Finally, the key information is effectively highlighted by performing dot-multiplication operations between the feature maps from the horizontal, vertical and temporal directions and the input feature maps. Experimental results on three typical public datasets show that the proposed method can better extract motion features and improve the accuracy of fall detection.