Dai Qionghai

Bio: Dai Qionghai is an academic researcher. The author has contributed to research in topics: Convolutional neural network & Data pre-processing. The author has an hindex of 4, co-authored 6 publications receiving 87 citations.

Image significance detection method based on confrontation network

Abstract: The invention discloses an image significance detection method which uses confrontation training to generate a convolution neural network model, which belongs to the field of computer vision and image processing The method comprises the steps of data preprocessing, network structure, suitable parameter selecting, and training with a random gradient descending method and an impulse unit According to data preprocessing, a large amount of collected data and labels are preprocessed According to network structure, a network structure and a specific kernel function are designed Suitable parameters including learning rate, a momentum factor and the number of images inserted into the network each time are selected The random gradient descending method and the impulse unit are used for training to reduce the possibility of network over-fitting According to the invention, a significance map can be accurately acquired

Depth map recovery method

Abstract: The invention discloses a depth map recovery method, comprising the following steps of A1, constituting a training set by the depth maps of a large number of various objects; A2, establishing a convolutional neural network (CNN), by using a nuclear separation method, acquiring the parameter of a hidden layer, establishing a convolutional network structure, and training the network structure and adjusting the network weight by using the depth maps in the training set; A3, in the output layer of the CNN, establishing an auto-regression model aiming at a possible result, and establishing an evaluation index; and A4, inputting an original depth map acquired by a depth sensor into the CNN, after denoising and classifying, recovering by an AR model, and if not conforming with requirements, inputting the result map into A2 until the high-quality depth map is acquired or the circulation is ended. According to the depth map recovery method, the image with low resolution and low signal to noise ratio acquired from the depth sensor can be recovered by using the depth convolution network. By using the depth map recovery method, the quality of the depth map can be significantly improved, and meanwhile the method for acquiring the depth map is also simplified.

Method for realizing super resolution for image

Abstract: The invention relates to a method for realizing super resolution for an image and belongs to the computer vision field. The method includes the following steps of: A1, data preprocessing: a certain number of high-resolution natural images are adopted to form a data set, a certain number of image blocks are extracted from the data set, Bicubic interpolation down-sampling and up-sampling in three times are carried out on the image blocks, and low-resolution images can be obtained; A2, network structure design: a designed convolutional neural network has 4 layers altogether; A3, hyper parameter selection: parameters such as network learning rate, learning momentum and batch_size are determined; and A4, network training and super parameter optimization: the convolutional neural network of all images in the training set from low-resolution images to corresponding high-resolution images is trained, and after any one image is inputted into the trained network, a high-resolution image can be obtained, so that the super resolution of the image can be realized.

Light field refocusing method

Abstract: The invention relates to a light field refocusing method, and the method comprises the following steps: initialization: extracting a sub-aperture image corresponding to an image photographed by a light field camera; recording position information: recording the position information of the extracted sub-aperture image of the light field camera; first super-recognition: taking one sub-aperture image according to the sequence, and carrying out the first super-recognition of the sub-aperture image through employing the trained super-recognition method; carrying out the super-recognition of the next sub-aperture image through employing the same method till the super-recognition of the last sub-aperture image is carried out; super-recognition focusing: obtaining a high-resolution refocus through employing the information of the sub-aperture images at the adjacent positions after refocusing conversion, a super-recognition reconfiguration method and a series of sub-aperture images. The multiple of a finally obtained light field camera super-resolution image is much greater than the multiple which can be obtained through a conventional super-recognition method, and greatly improves the resolution of a light field camera image obtained through a conventional method.

Significance detection method of light field image

Abstract: The invention discloses a significance detection method of a light field image. The method comprises the following steps of S1, carrying out refocusing on different positions of the light field image to acquire N focusing images, and fusing the N focusing images to acquire a full-focusing image, wherein the N is a positive integer; S2, calculating a focusing degree F(x, y) of each pixel point of each focusing image, wherein the x and the y represent an abscissa and an ordinate of the pixel point (x, y) respectively; Gx and Gy represent gray gradient values of an x direction and a y direction of the pixel point (x, y) respectively; S3, calculating one-dimensional focusing degree distribution of each focusing image; S4, calculating a background layer approximation degree of each focusing image; and S5, taking the focusing image corresponding to a maximal background layer approximation degree BLS(Ii) as a background layer of the full-focusing image. In the invention, a background can be accurately detected and separated; processing of color contrast and the like can be performed on foreground information so as to acquire an accurate saliency map.

Pest and disease image generation method based on generative adversarial network

Abstract: The invention relates to a pest and disease image generation method based on a generative adversarial network In the prior art, sampling images of pest and disease images are less By using the method of the invention, the above defect is overcome The method comprises the following steps of collecting and preprocessing trained images; based on a deep-convolution neural network model, constructing a discrimination network and a generation network; training the discrimination network and the generation network; and according to the trained generation network, generating the pest and disease images In the invention, according to a few of existing pest and disease images, a lot of pest and disease images which are similar to a reality are generated, a sample image is provided for pest and disease image identification, and problems that the pest and disease images in an actual field are less and acquisition cost is high are solved

Depth significance-based remote sensing image rapid retrieval method

Abstract: A depth significance-based remote sensing image rapid retrieval method is disclosed and belongs to the field of computer vision. The method disclosed in the invention specifically relates to technologies such as in-depth learning, significance object detection, image retrieval and the like. According the method, remote sensing images are research objects, and in-depth learning technologies are used for researching a remote sensing image rapid retrieval method. A full convolution neural network is adopted for constructing a multitask significance object detection model which is used for doing significance detection tasks and semantic segmentation tasks at the same time, and depth significance characteristics of the remote sensing images are learnt in network pre-training processes. A depth network structure is improved, a Hash layer fine tuning network is added, and binary system Hash codes of the remote sensing images can be obtained via learning. Significance characteristics and the Hash codes are used comprehensively for similarity measurement. The method disclosed in the invention is of high application value for realizing accurate, highly efficient and feasible retrieval of the remote sensing images.

Monocular image depth estimation method based on multi-scale CNN and continuous CRF

Abstract: A monocular image depth estimation method based on multi-scale CNN and continuous CRF In the method, a CRF model is used to calculate single-point potential energy according to an output depth map of DCNN, paired sparse potential energy is calculated according to an input RGB image, and finally an optimized depth map is deduced using a maximized posterior probability (MAP) algorithm According to the method, optimized thoughts of multi-scale CNN and continuous CRF are combined, so that not only a depth map can be estimated with a relatively high precision, the contour of the obtained depth map can also be clear The depth estimated by means of the method has a relatively high resolution, and the obtained depth map can reserve depth detail information about all objects in a scenario, thereby having a better visual effect

Identification method of emotional tendency of network comment texts and convolutional neutral network model

Abstract: The invention discloses an identification method of emotional tendency of network comment texts and a convolutional neutral network model. The method comprises the steps as follows: grabbed network comment texts constitute a data set; word segmentation and text preprocessing are performed; all words subjected to text preprocessing are trained, and word vector representation of all words is obtained; the convolutional neutral network model is constructed and trained on a training set selected from the data set, and network parameters are updated with a back-propagating algorithm; in each step of training, noise is added to word vectors of an input layer for construction of adversarial samples, adversarial training is performed, and network parameters are updated with a random gradient descent algorithm; a classification model is obtained through repeated iteration to identify the emotional tendency of the network review texts. The convolutional neutral network model is used in the method and comprises the input layer, a convolution layer, a pooling layer and a classification layer. The adversarial samples can be classified correctly and the identification accuracy is improved.

Video enhancement and transmission method based on deep learning

Abstract: The invention discloses a video enhancement and transmission method based on deep learning. Downsampling is carried out on a high-definition video at a video source end, thereby obtaining a low-definition video; the low-definition video is compressed in an existing video coding mode; and the compressed low-definition video is transmitted. The size of the video data is greatly reduced through downsampling and video coding, so the video traffic needing to be transmitted is correspondingly reduced, and an effect of reducing the bandwidth cost is achieved. At a user receiving end, a user receives the low-definition video, reconstructs the low-definition video by employing a super-resolution image reconstruction method of the deep learning and restores the low-definition video into a high-resolution video for the user to watch, so the video transmission bandwidth cost is effectively reduced. According to the method, the video transmission bandwidth cost is reduced by at least 50%; the resolution of the live broadcast video is improved; and the watching experience of a user is improved.