Nanjing University of Posts and Telecommunicatins

Other

Dengyin Zhang

Weidan Yan

Junjiang Li

Can Hua Chen

Chao Zhou

Xiaofei Jin

Shasha Zhao

Zijie Liu

Qunjian Du

Yepeng Xu

Chenghui Qi

Wensheng Han

Rong Zhao

Yan Yang

Min Zhang

Hong Zhu

Qian Zhao

Lin Zhu

Fei Ding

Fan Gao

Yi Cheng

Yuanpeng Zhao

Yu Qiu

Jingyu Wang

Nianqi Zhang

Yingjie Kou

Zheng Zhou

Xu Li

Wenjie Zhang

Siyu Wang

Jie Dong

Yongjun Li

Can Chen

Junjie Sun

Chen Yang

Kang Liu

Lidan Qin

Shiqi Zhou

Jiawei Gu

Shuying Xu

Yi Xiao

Jia Zhong

Lufeng Yuan

Wenye Ni

Yingying Feng

Wenhong Xin

Songhao Lu

Yonglian Ma

Yingying Cao

Zhaoxi Shi

Rui Gong

Yingying Zhou

Jinshuai Wang

Chenhui Sun

Shuo Yang

Ruijie Chen

Xuejie Cao

Shibo Kang

Yue Cai

Xiaoye Ruan

Zhou Ye

Maomao Ji

Ying Zhao

Chengwan Ai

Qing Wen

Jiangwei Dong

Yufan Liu

The present invention pertains to the technical field of smoke detection and discloses a lightweight fire smoke detection method, a terminal device, and a storage medium. Primarily, a smoke-like foreground is screened out based on a combination of a mixed Gaussian model and a YUV color model; and an ACON activation function is introduced to replace a Leaky ReLU activation function of YOLOv5 to form an ACON-CSP module for feature extraction. In the present invention, the smoke foreground is extracted for preprocessing by using the combination of the mixed Gaussian and YUV. Thus, in the preprocessing stage, static interference and non-smoke color interference in the preprocessing stage are ruled out while narrowing down the range of smoke detection, ensuring the relative accuracy and improving the detection speed, thereby providing a possible method for fire smoke detection of a low-end terminal device in an outdoor computer.

Lightweight fire smoke detection method, terminal device, and storage medium

The present disclosure discloses a small target detection method based on improved YOLOv8, including: inputting a small target image to be detected into a pre-trained small target detection model based on the improved YOLOv8 for identification to obtain a detection result, where a method for training a small target detection model based on the improved YOLOv8 includes: acquiring a small target image data set and dividing the small target image data set into a training set and a validation set; replacing a backbone network of YOLOv8 with a backbone network ATDeNet and constructing the small target detection model based on the improved YOLOv8; and training the constructed small target detection model by using the training set and the validation set to obtain a trained small target detection model based on the improved YOLOv8. The accuracy and efficiency of small target detection can be significantly improved.

Small target detection method based on improved YOLOv8

The present invention discloses a method and an apparatus for obstacle detection under complex weather. The method includes: obtaining an image under a complex weather condition; performing enhanced preprocessing on the image by using a multi-scale retinex with color restoration MSRCR algorithm; inputting the preprocessed image into a trained obstacle detection model based on an improved YOLOv3 network; and according to output of the obstacle detection model based on the improved YOLOv3 network, determining an obstacle detection result under the complex weather; replacing a Leaky-ReLU activation function in convolutional layers in the original YOLOv3 network with an ELU activation function; and training the obstacle detection model with the processed data set to obtain a trained obstacle detection model based on the improved YOLOv3 network.

Method and apparatus for obstacle detection under complex weather

The present application discloses a task scheduling method based on an improved particle swarm optimization algorithm, which includes: obtaining task data to be scheduled, encoding particles according to the task data; iterating the particles by a particle swarm optimization algorithm; in response to that the particle swarm optimization algorithm does not fall into a local optimal solution, outputting a scheduling scheme; and in response to that the particle swarm optimization algorithm falls into the local optimal solution, outputting the scheduling scheme by fusing the particle swarm optimization algorithm with a cuckoo search algorithm. The present application introduces a cuckoo search algorithm when the particle swarm optimization algorithm falls into a local optimal solution, solving the dilemma of the particle swarm optimization algorithm falling into a local optimal solution, while improving the global search capability of the algorithm.

Task scheduling method based on improved particle swarm optimization algorithm

Disclosed is a multi-scale fusion defogging method based on a stacked hourglass network, including inputting a foggy image into a preset image defogging network; and outputting a fogless image after the foggy image is processed by the image defogging network. The image defogging network includes a 7×7 convolutional layer, a stacked hourglass module, a feature fusion, a multi-scale jump connection module, a 1×1 convolutional layer, a 3×3 convolutional layer, a hierarchical attention distillation module, the 3×3 convolutional layer and the 1×1 convolutional layer connected sequentially.

Multi-scale fusion defogging method based on stacked hourglass network

The present disclosure discloses a person re-identification method of integrating global features and ladder-shaped local features and a device thereof, comprising: extracting the image to be identified and the image-library image by using a pre-trained person re-identification network model, respectively, to obtain the person feature of the image to be identified and the person feature of the image-library image; matching the similarity between the person feature of the image to be identified and the person feature of the image-library image, and outputting the person images with the top N similarity as the person re-identification result; wherein a person re-identification network comprises a backbone network, an improved global feature branch and a ladder-shaped local feature extraction branch guided by a block weight; using a public data set to train the person re-identification network, and obtaining the trained person re-identification network model.

Person re-identification method of integrating global features and ladder-shaped local features and device thereof

An image defogging method based on dark channel prior is provided, including: acquiring a dark channel image and a light channel image of a haze weather image according to a haze weather image; selecting pixel values in the light channel image and calculating an atmospheric light value A; obtaining a transmittance image based on the light channel image and the atmospheric light value A; obtaining a restored image of defogging based on the transmittance image and the atmospheric light value A. By using the pixel values of the light channel to distinguish whether pixel points comply with the dark channel prior, calculating the transmittance of image areas that comply with the dark channel prior and those that do not, so as to avoid the situation where the image defogging effect is not good due to inaccurate transmittance estimation, making the defogging of images containing high brightness areas more clear.

Method for image defogging based on dark channel prior

The present application discloses a method for recovering a face image based on semantic features, which includes obtaining a to-be-recovered face image; inputting the to-be-recovered face image into a recovery model trained to obtain a recovered face image; the recovery model includes: an encoder, configured for generating low-quality face semantic features based on the to-be-recovered face image; a reference image generator, configured for generating a plurality of high-quality face reference images by inputting random noise based on the low-quality face semantic features; a feature transfer, configured for constructing a face component feature dictionary based on the plurality of high-quality face reference images, and transferring high-quality component features in the face component feature dictionary to the low-quality face semantic features to obtain high-quality face semantic features; and a decoder, configured for generating the recovered face image based on the high-quality face semantic features.

Method for recovering face image based on semantic features

The present disclosure discloses a multiscale point cloud classification method. The method includes the following steps: acquiring 3D unordered point cloud data; performing feature extraction and classification on the acquired point cloud data using a pre-trained parallel classification network to obtain an output result, wherein the parallel classification network includes a plurality of basic networks with the same structures; and fusing the output results of the parallel network using a pre-trained deep Q network to obtain a final result of point cloud classification. The present disclosure can improve the accuracy and robustness of point cloud classification.

Multiscale point cloud classification method and system

The present application discloses a face image restoration method based on a state space model, which includes inputting the to-be-restored face image to a restoration model to obtain a restored face image; the restoration model includes an encoder and a decoder; the encoder sequentially includes a first image fusion module, a first multi-scale state space module, a second image fusion module, a second multi-scale state space module, a third image fusion module, and a third multi-scale state space module; the decoder sequentially includes a fourth multi-scale state space module, a first multi-scale attention fusion module, a fifth multi-scale state space module, a second multi-scale attention fusion module, a sixth multi-scale state space module, and a third multi-scale attention fusion module. This method ensures the consistency of face semantic information while restoring the detailed texture, especially low quality face images in real degraded scenes can achieve better restoration results.

Face image restoration method based on state space model

The present application discloses a batch-type image distribution method based on an IPFS, which includes: obtaining a list of cluster nodes needed to download images; calculating a caching ratio of an image layer of images to be distributed of each node, comparing the caching ratio of each node with a threshold of a preset node, and obtaining a complete image from an image repository; adding the node with the complete image to a list of source nodes, and adding remaining nodes needed to download images to a list of demand nodes, and constructing an IPFS network, and exporting and sharing the image; calculating target demand nodes using an optimization algorithm, and inputting the target demand nodes into the IPFS network for image download, and importing the downloaded image into a container and starting the container.

Batch-type image distribution method based on IPFS, system and storage medium

Disclosed are an indoor non-contact human activity recognition method and system. The method comprises: collecting an indoor reflected signal by using an antenna array; filtering the reflected signal to obtain a noise-removed reflection signal; and inputting the noise-removed reflected signal to a pre-trained human activity recognition model, and determining a human activity category, the human activity recognition model being a pre-trained CNN network model based on a transfer learning algorithm. The recognition method and system have the advantages that: the antenna array is configured for collecting human actions to carry out activity recognition indoors, which can be applied to home-based care scenes; original data is denoised, so that most of high-frequency noises can be removed, and a phase change of the signal is reserved; a CNN structure is adopted for training so as to reduce a complexity of the system location-free sensing.

Indoor non-contact human activity recognition method and system

The present application discloses a method and a device for user grouping and resource allocation in a NOMA-MEC system. The hybrid deep reinforcement learning algorithm proposed in the present application solves the hybrid problem of deep reinforcement learning that is difficult to deal with both discrete and continuous action spaces by using DDPG to optimize continuous actions and DQN to optimize discrete actions. Specifically, the algorithm determines a bandwidth allocation, an offloading decision, and a sub-channel allocation (user grouping) of the user device based on the user's channel state, in order to maximize the ratio of the computation rate to the consumed power of the system. The algorithm is well adapted to the dynamic characteristics of the environment and effectively improves the energy efficiency and spectrum resource utilization of the system.

Method and device for user grouping and resource allocation in NOMA-MEC system based

Disclosed are an image dehazing method and system based on CycleGAN. The method comprises: acquiring a to-be-processed hazy image; and inputting the image into a pre-trained densely connected CycleGAN, and outputting a clear image. The densely connected CycleGAN comprises a generator, the generator comprises an encoder, a converter and a decoder, the encoder comprises a densely connected layer for extracting features of an input image, the converter comprises a transition layer for combining the features extracted at the encoder stage, the decoder comprises a densely connected layer and a scaled convolutional neural network layer, the densely connected layer is used for restoring original features of the image, and the scaled convolutional neural network layer is used for removing a checkerboard effect of the restored original features to obtain a finally output clear image.

Image dehazing method and system based on cyclegan

A medical image segmentation method include: 1) acquiring a medical image data set; 2) acquiring, from the medical image data set, an original image and a real segmentation image of a target region in the original image in pair to serve as an input data set of a pre-built constant-scaling segmentation network, the input data set including a training set, a verification set, and a test set; 3) training the constant-scaling segmentation network by using the training set to obtain a trained segmentation network model, and verifying the constant-scaling segmentation network by using the verification set, the constant-scaling segmentation network including a feature extraction module and a resolution amplifying module; and 4) inputting the original image to be segmented into the segmentation network model for segmentation to obtain a real segmentation image.

Medical image segmentation method based on U-network

Growing community of inventors

Nanjing, China

Dengyin Zhang