The invention discloses a cascaded residual error neural network-based image denoising method. The method comprises the following steps of building a cascaded residual error neural network model, wherein the cascaded residual error neural network model is formed by connecting a plurality of residual error units in series, and each residual error unit comprises a plurality of convolutional layers, active layers after the convolutional layers and unit jump connection units; selecting a training set, and setting training parameters of the cascaded residual error neural network model; training the cascaded residual error neural network model by taking a minimized loss function as a target according to the cascaded residual error neural network model and the training parameters of the cascaded residual error neural network model to form an image denoising neural network model; and inputting a to-be-processed image to the image denoising neural network model, and outputting a denoised image. According to the cascaded residual error neural network-based image denoising method disclosed by the invention, the learning ability of the neural network is greatly enhanced, accurate mapping from noisy images to clean images is established, and real-time denoising can be realized.

Cascaded residual error neural network-based image denoising method

The invention discloses a medical image classifying method based on image feature migration learning and deep learning. According to the medical image classifying method, features of a medical image are extracted by virtue of a trained model which well presents in the image understanding field, and an obtained feature graph is used for training a convolutional neural network model capable of realizing final classification; the features extracted from the existing model which well presents in the image understanding field are adequately utilized, and a relatively good effect can be obtained from a small-scale medical image data set.

Medical image classifying method

The invention discloses a method for effectively enhancing the robustness of a convolutional neural network The method comprises the steps that random two-dimensional transformation is performed on an input characteristic graph in forward transmission and then convolutional operation is performed, and finally the prediction value of the classification result of the input characteristic graph is obtained; backward transmission is performed on the error between the obtained prediction value and a real value in backward transmission, the corresponding gradient of the parameters of each layer is calculated and then the parameters are updated along the reverse direction of the gradient; the forward transmission process and the backward transmission process are repeated for the preset number of times and then a model of the optimal parameters is obtained; and the model of the optimal parameters is tested in testing so that the robustness of the convolutional neural network can be enhanced When the method is used for image classification and retrieval, layer-by-layer random two-dimensional transformation is introduced in the training process with no requirement for introducing a new characteristic extraction module, new parameters or performing extra processing on the input image so that the robustness of the convolutional neural network can be effectively enhanced and the complexity of image processing can be reduced

Method for effectively enhancing robustness of convolutional neural network

Provided is a pearl classification method based on depth learning The method includes the steps of 1) acquiring the pearl images as sample data; 2) preprocessing the sample data; 3) dividing the sample data into training data and test data; 4) using the training data to train the deep convolution network; 5) extracting the features of the training data and the test data by means of the trained deep convolution network; 6) constructing an SVM classifier by means of the features of the training data extracted in step 5) and classifying the test data by means of the SVM classifier The depth learning is carried out by means of the multiple views of pearls, the self learning advantages of depth learning can be fully played, the good characteristics can be automatically learned, the tedious manual extraction of features and design features can be prevented, and moreover, the classification of the pearls is carried out through the combination of the SVM, and whether the pearls have threads can be effectively determined

Pearl classification method based on depth learning

The invention discloses a method and a system for classifying multiple objects in an image and a computer system. The method comprises the steps of filtering the image to generate a first feature pattern, wherein the first feature pattern comprises first feature values used for representing semantic characteristics of the objects; generating a plurality of second feature patterns according to the first feature pattern and a plurality of bias values, wherein each second feature pattern comprises a plurality of second feature values corresponding to the first feature values which are subjected to bias processing through a bias value respectively; updating each second feature pattern to enhance the difference among the second feature values in each second feature pattern; and classifying the objects according to the updated second feature patterns. According to the scheme provided by the method and the system, the requirement on the number of filters used for image filtration is lowered, the calculation burden is reduced, and the image identification speed is increased.

Method and system for classifying multiple objects in image and computer system

The invention relates to a hyperspectral data classification method based on a multi-layer convolution network and data organization and folding. The method comprises: step one, pretreatment is carried out before expanding and classification of three-dimensional hyperspectral data and a data matrix including effective spectrum information and a tag vector are obtained; step two, feature dimension expanding is carried out on the data matrix, and column-based folding and reorganization are carried out on a feature dimension to obtain a reorganized three-dimensional hyperspectral data input matrix; step three, a multi-layer convolution network structure parameter and an initial value are set; and step four, a feature and an error are calculated layer by layer by using forward propagation and BP algorithms, a network weight and a bias value are updated, iteration is carried out continuously to obtain a network stablity parameter, and then a network model for classification and a parameter for classification are obtained. Compared with other methods, the provided method has advantages of clear principle, clear structure, short identification time, and high detection identification rate; and the method being an effective classification method for hyperspectral data is suitable for rapid target detection and classification identification application of hyperspectral images.

Hyperspectral data classification method based on multi-layer convolution network and data organization and folding

The invention discloses a hyperspectral unmixing algorithm based on a de-noising three-dimensional convolutional self-encoding network, and solves the problems that the traditional end member extraction algorithm cannot quickly and synchronously obtain end member spectrum and abundance information, is easily interfered and is poor in robustness. The method comprises the following steps: 1, designing and establishing a three-dimensional convolutional self-encoding network based on a hyperspectral volume data structure; 2, training a network by using a denoising self-encoding algorithm and a generated data set, and extracting robust space-spectrum joint features; and 3, designing and realizing a non-negative sparse auto-encoder, adding non-negative and sparse constraints to abundance, and mapping data to a low-dimensional signal subspace so as to synchronously obtain high-precision end member and abundance information. The basic idea of the method is that the deep self-encoding network is adopted to extract the spatial-spectral joint features, the end member and abundance information of the image is synchronously obtained in an unsupervised mode, the unmixing precision is high, the method is suitable for hyperspectral end member extraction and abundance inversion application such as urban remote sensing, precision agriculture and exploration and investigation, and the social andeconomic value is high.

Hyperspectral unmixing algorithm based on denoised three-dimensional convolutional self-encoding network

The invention discloses a linear traveling line spectral data object detection method based on a difference and a convolution kernel, comprising steps of performing high spectral data pre-processing, removing exaggeration wave bands to perform de-equalization processing, choosing a suitable step length to perform traveling line sampling on the data, performing difference operation and convolution operation to obtain a characteristic value measuring the sampling cuboid, choosing a suitable threshold characteristic value to perform processing, wherein the area exceeding the threshold contains an object point, repeating the step 2 and the step 3 until traversing a whole cuboid data set to find the transverse position information of the object point, performing transposition on the hyperspectral data obtained from the step 1 in an image dimension, repeating the step 2 to step 4 until traversing the whole cuboid data set to find the column position information of the object point, and thus positioning the position information of the object point in the image. The method is clear in pricinple, clear in structure, small in calculation, low in time complexity, high in detection recognition and is applicable to the high speed object detection and positioning application of the high spectral image under the uniform background.

Shen Yi

Papers

Hyperspectral data classification method based on multi-layer convolution network and data organization and folding

Hyperspectral unmixing algorithm based on denoised three-dimensional convolutional self-encoding network

Traveling line sampling high spectral data object detection method based on difference and convolution kernel