37results about How to "Increase the amount of training data" patented technology

The invention discloses an industrial product surface defect detection method based on sample enhancement. The method comprises the following steps: 1) carrying out size standardization, normalization, cutting and classification on an industrial product surface image; 2) carrying out data enhancement of random flipping on the picture with the defect; 3) randomly splicing and enhancing the defective picture and the normal picture; 4) carrying out iterative training by using a Cascade-RCNN algorithm; 5) acquiring a Cascade-RCNN detection model; 6) carrying out sliding window detection on the industrial product surface picture to be detected and the texture template picture determining no defect through the Cascade-RCNN detection model, splicing results detected by a sliding window, and comparing the results obtained by the sliding window detection and the texture template picture to finally obtain the defect category and region annotation of the picture to be detected. According to the method, the influence of conditions such as illumination, exposure and displacement on defect detection can be effectively reduced, the detection stability is improved, the resolution capability of thetwo-stage target detector on patterns and backgrounds is improved, and the false detection rate is reduced.

The invention discloses a face data enhancement method based on key point disturbance technology, comprising steps of utilizing a face calibration method based on the human face key point technology, performing disturbance on the key point position, and obtaining a new human face image through linear transformation, affine transformation, and partitioning affine transformation . The human face calibration method based on the human face key point technology performs the human face calibration by the linear transformation, the affine transformation, and the partitioning affine transformation, performs disturbance on the position of the key point, and enabling one human face image to generate a plurality of human face images. The face data enhancement method based on the key point disturbance technology can solve the problem that the data is insufficient in the process of human face recognition training, generates the plurality of images through one human face image, increases the training quantity and improves the accuracy of the human face recognition.

The invention provides a method and a system for performing full-automatic segmentation on multiple tissues of an oral cavity CBCT image. According to the obtained original oral cavity three-dimensional CBCT data, a deep learning method is utilized to label the data, a deep learning method is utilized again to segment a new image, the image processing process is simple, excessive manual labeling and intervention are not needed, the difficulty of existing medical data annotation is overcome, and meanwhile human differences in the image annotation process are reduced. In the network training process, the adversarial network is generated for the training data, the training data volume is increased, and the universality of the network can be effectively improved. The establishment of the method greatly promotes the segmentation of a medical image technology, and provides an effective technical support for the automatic diagnosis and prognosis of oral diseases.

The invention discloses a transaction object prediction method and device, and relates to the technical field of computers. One specific embodiment of the method comprises the steps of obtaining a plurality of prediction reference texts of a target transaction object in each period of a plurality of historical transaction periods, and determining a fluctuation tendency factor of each prediction reference text; determining a text tendency index of the target transaction object in the period by utilizing the rising and falling tendency factor; and inputting the price data and the text tendency index of the target transaction object in each period of the plurality of historical transaction periods into a pre-trained prediction model to obtain information of at least one preset dimension of the target transaction object in a future transaction period. According to the embodiment of the invention, the method can improve the prediction accuracy of the transaction object through introducing the features, affecting the change trend of the transaction object, in the prediction reference text into the model for training and prediction and analyzing a large amount of historical data containing the features.

The invention discloses a 3D point cloud semantic segmentation and migration method based on meta-learning, and relates to the technical field of robot navigation. The method comprises the following steps: constructing a Point Net network model; selecting a training data set; for each training data set, forming a training task set by utilizing different types of data; constructing a meta-learningframework; according to the meta-learning framework, training the Point Net network model through each training task set; selecting a test task set; and inputting the test task set into the trained Point Net network model for testing until the gradient update value of the model converges. According to the method, a trained model is loaded in a new environment task, optimal similar task parametersare used, and the training efficiency of the indoor scene semantic segmentation method of the new task is high; the semantic segmentation capability of different tasks is learned through a meta-learning framework, it is ensured that learning features can be suitable for different migration environments, and the generalization performance of the model is improved.

The invention discloses an optic nerve simulation method and an optic nerve simulation system based on deep learning, wherein the method comprises the steps of transmitting a visible light signal; acquiring a visual neural electrical signal for obtaining a first visual neural simulated electrical signal, performing two-dimensional imaging on a brain visual processing cortex for obtaining a first near-infrared brain functional area simulated two-dimensional image; performing filtering, prior modification and sampling on the received visible light signal, and furthermore obtaining a second visual neural simulated electrical signal and a second near-infrared brain functional area simulated two-dimensional image; and performing analysis and evaluation, using data after evaluation as parameters for correcting a deep leaning module, and transmitting the data to the deep leaning module in a multi-dimensional vector form, thereby reconstructing a deep learning network by means of the multidirectional vector. According to the optic nerve simulation method, medical simulation and rehabilitation training can be performed on a human brain and an optic nerve system by means of a deep learning method. Simple operation and high reliability are realized. Furthermore along with increase of training data volume, system gain can be remarkably improved.

The invention discloses a method for generating large batch of point cloud data sets. The method comprises the following steps: establishing a three-dimensional model for an object in a scene; re-sampling points for each vertex and each patch in the three-dimensional model to form a surface point cloud of the three-dimensional model; combining the surface point clouds of components in the scene togenerate scene point cloud information; processing the scene point cloud information by adopting a simulated camera model to form simulated scene point cloud information acquired by a simulated structured light system; generating a large batch of point cloud data sets by repeating for many times. According to the method, the training data volume of a neural network for point cloud processing is improved, and meanwhile, the method has pertinence for specific scenes and processing tasks, so that the performance of the neural network is improved.

The embodiment of the invention relates to the field of face recognition, and discloses a model training method, a relative relationship recognition method, electronic equipment and a storage medium. The model training method comprises the following steps: constructing a family face feature extraction model for extracting family face features from family image samples; wherein the family image sample comprises a plurality of face images belonging to the same family, and the family face feature extraction model is provided with organ parameters for fusing individual face features of the plurality of face images to form family face features; the organ parameter represents the dominant degree of the face feature of the corresponding organ region in the individual face feature in the family face feature; and constructing a loss function according to the distance between the family face feature of each family and the corresponding family label and the distance between the family face feature and the face feature of each face image in the corresponding family image sample, and training a family face feature extraction model.

The invention discloses a convolutional neural network training method, an electroencephalogram signal recognition method and device and a medium, and the method comprises the steps: executing a plurality of obtaining processes, obtaining an electroencephalogram signal in each obtaining process, and executing the time domain data enhancement and frequency domain data enhancement of the electroencephalogram signal, and training a convolutional neural network by using the enhanced electroencephalogram signal, and the like. The convolutional neural network trained by the method is a multi-input,multi-convolution-scale and multi-convolution-type hybrid convolutional neural network, the sizes of a multi-input convolution layer and a convolution kernel are reasonably designed, and the method has high recognition accuracy; a training set used for training the convolutional neural network is obtained by performing time domain data enhancement and frequency domain data enhancement expansion based on the acquired electroencephalogram signals, so the training data volume of the convolutional neural network can be increased, the over-fitting phenomenon can be reduced, noise interference in the electroencephalogram signals can be effectively coped with, and the recognition effect can be improved. The method is widely applied to the technical field of signal processing.

Embodiments of the invention disclose a deep learning image registration method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining deformation fields ofa first fixed image and a moving image through a first algorithm, acquiring a second fixed image according to the deformation field and the moving image, and training a deep neural network accordingto the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. According to the deep learning image registration method provided by the embodiment of the invention, the simulated deformation field is generated by using the generation algorithm and is not artificially labeled, so that the registration network is better trained, and the problems that the registration standard is defined and the registration process is driven by depending on the artificially set registration metric in the prior art are solved. According to the method, the training data volume is greatly improved in the neural network, the registration accuracy is further improved, and the calculation speed is high.

The invention discloses a lung pathological sound automatic analysis method based on multi-task classification, and relates to the technical field of lung pathology analysis. The method comprises the following steps: inputting extracted audio features into a multi-task classification model of a convolutional neural network MobileNetV2, wherein the multi-task classification model of the convolutional neural network comprises the steps of outputting a lung pathological sound recognition task and outputting a lung illness prediction task. According to the method, the training data volume can be implicitly increased by adopting a multi-task learning method, the generalization performance of the model is improved through domain knowledge of multiple pieces of label information of the same data, so that the prediction accuracy of the multi-task classification model of the convolutional neural network MobileNetV2 is improved, and in addition, the lightweight multi-task classification model of the convolutional neural network MobileNetV2 is adopted, so that the number of parameters is small, the requirements for the computing power and the memory size of training equipment are small, and the prediction classification task can be completed on mobile or embedded equipment.

The invention relates to a machine learning model training method and device and an image classification method and device, and relates to the technical field of artificial intelligence. The training method comprises the steps that a first machine learning model is utilized to extract a feature vector of a to-be-processed picture, a first classification result of the to-be-processed picture is determined according to the feature vector, and the to-be-processed picture belongs to a first data field or a second data field; according to the feature vector, a second machine learning model is utilized to determine a second classification result of the to-be-processed picture, and the second classification result comprises a classification result of the to-be-processed picture in the first data domain and a classification result of the to-be-processed picture in the second data domain; and according to the first classification result and the second classification result, adversarial training is performed on the first machine learning model and the second machine learning model, so that the accuracy of the second classification result is lower than a threshold, and the trained first machine learning model is used for picture classification.

The invention discloses an NLP-based listed company financial fast message automatic generation method and system in the technical field of natural language processing, and the method comprises the steps: carrying out the structural processing of an announcement file, and obtaining text analysis information; extracting financial data and text data from the text analysis information to obtain key financial information; and inputting the key financial information into the trained improved seq2seq model to generate a financial message. The method has the characteristics that the generated financial message text is rich in style, simple to maintain and use, good in text matching effect and the like.