36results about How to "Reduce incompleteness" patented technology

The invention relates to the technical field of ECG analysis and in particular relates to an ECG signal classification method and device. The ECG signal classification method comprises the following steps: a, segmenting the ECG signal, and respectively acquiring training set data and test set data; b, performing model training on the training set data through deep learning, and constructing a prediction classification model; and c, performing ECG signal classification on the test set data through the prediction classification model. According to the embodiment of the invention, the training set is subjected to model training through deep learning so as to obtain the prediction classification model, and the ECG signal classification is performed through the prediction classification model, so that incompleteness caused by artificial design characteristics is reduced, the ECG signal classification accuracy is improved, and more types of heart rate variability can be classified.

The invention belongs to the technical field of computer vision, and discloses a convolutional neural network-based wind driven generator blade crack detection method, which comprises the following steps of (1) acquiring a generator blade image, and constructing a learning model training sample; (2) training a convolutional neural network model by using the model training sample; (3) preprocessingthe to-be-detected image; (4) carrying out feature extraction on the image by adopting a feature extraction network to obtain a feature map; (5) inputting the feature map into an area generation network to obtain a blade existence probability in each candidate box and the candidate box initial position coordinates; (6) performing the threshold filtering and non-maximum suppression on the candidate box; (7) inputting the feature map of each candidate box region into an interested region pooling layer and a frame regression network to obtain the candidate box correction coordinates; and (8) inputting the original image areas corresponding to the candidate frames into a classification network, and judging a blade crack classification result. According to the method, the interference of the image background content is eliminated, and the blade detection precision is improved.

The invention relates to a wave water surface sun glitter stripping method based on polarization remote sensing detection, comprising the following four steps: step one, selecting proper time and point to carry out a water surface polarization remote sensing test; step two, selecting proper instruments to acquire a wave water surface sun glitter area time sequence polarization image; step three, processing a wave water surface sun glitter area polarization image; and step four, stripping wave water surface sun glitter. The method is a convenient and efficient wave water surface sun glitter correction method, and has the advantages of being high in precision and good in real-time without causing image data loss or relying on detection angle selection, and the like; and the method has extremely important application values in the fields of dynamic monitoring of sea surface ships, underwater target detection, water color remote sensing, and the like.

The invention discloses a frozen fresh pomfret freshness rapid grading method based on deep learning, which belongs to the technical field of refrigerated food quality evaluation. Specifically, freshness grading is carried out on a pomfret image based on a Mask R-CNN technology, a frozen fresh pomfret image is obtained, a digital image data set is established, feature learning is fused into the process of establishing the model, accurate detection and segmentation are carried out on fisheye and gill parts, feature learning can be carried out autonomously, and the freshness of the pomfret can be accurately segmented and classified. Therefore, quick detection and intelligent prediction of the quality deterioration of the frozen fresh pomfret are realized; and the expandability and mobility of the model are improved. A rapid, low-cost, accurate, nondestructive and real-time automatic detection technology is realized, and the health and benefits of consumers are guaranteed. Product qualityis controlled for different varieties of cold chain storage and transportation environments and supply chain sources, and technical support is provided for public health and epidemic prevention.

An automatic-drawing device of article title and correlation information consists of title candidate sentence drawing unit for drawing out multiple title candidate sentence from test article inputted by article input unit, characteristic value drawing unit for drawing out characteristic value from each of multiple title candidate sentence and title deciding unit of deciding out title from multiple title candidate sentence according to drawn out characteristic value.

The invention provides a water environment data transmission system and method based on Internet of Things, belonging to the technical field of Internet of Things. The water environment data transmission system comprises a collection module, a protocol analysis and conversion module, a data processing module, a data transmission management module, a database and an application layer module. The method comprises a collecting step, a protocol analyzing and converting step, a data processing step, a data transmission management step, and a database and application layer step. The water environment data transmission system and method can be applied to various water environment monitoring systems, and have the advantages of real-time data monitoring, high accuracy and the like.

The invention discloses an electrocardiogram feature extraction method, device, system, equipment and classification method based on a deep learning algorithm. The electrocardiogram feature extractionmethod based on the deep learning algorithm comprises the following steps: randomly intercepting a section of continuous electrocardiogram signal from a twelve-lead electrocardiogram to be processed,wherein the electrocardiogram signal at least comprises two cardiac cycles; and inputting the intercepted electrocardiogram signal into a feature extraction model in a picture form, and performing extraction to obtain electrocardiogram signal features, wherein the feature extraction model is obtained by training a ResNet model or an Inception model or an Inception-ResNet model. According to the electrocardiogram feature extraction method, device, system, equipment and classification method based on the deep learning algorithm provided by the invention, the incompleteness caused by artificialdesign features is reduced, and the accuracy and diversity of electrocardiogram feature extraction based on the deep learning algorithm are improved.

The invention discloses a method for evaluating the performance of cultivated land requisition-compensation balance based on a PSR model. The method includes the steps of establishing a correspondingpressure state response (PSR) model index system for each evaluation index element under a cultivated land requisition-compensation balance policy; adopting a TOPSIS method to calculate a pressure index evaluation value, a state index evaluation value and a response index evaluation value under the PSR model index system; adopting a coordination index function to calculate a coordination degree value between the pressure index evaluation value, the state index evaluation value and the response index evaluation value. Through the method, the PSR model embodies the interaction between humans andthe environment, emphasizes the influence of population, society, and economy on policies, and can comprehensively evaluate the whole process of policies, effectively making up for the above defects..

The invention provides a behavior recognition method based on three descriptors of hydrodynamics. The behavior recognition method based on three descriptors of hydrodynamics comprises the steps: calculating three descriptors of a picture, based on hydrodynamics, and respectively obtaining an advection descriptor, a diffusion descriptor and a pressure descriptor of the picture; integrating thermodynamic diagrams of different descriptors of the same frame into the input data type of the adopted neural network, inputting into the trained neural network, performing fine tuning on the neural network to obtain a model of different crowd behaviors; and by classifying the crowd behaviors through the obtained model, obtaining the detected crowd behavior type, and realizing behavior recognition. Thebehavior recognition method based on three descriptors of hydrodynamics has high accuracy for behavior recognition, can reduce a frame in which no event occurs as a frame in which an event occurs, and can also detect more frames in which events occur.

Provided are an image processing apparatus and an image processing method that can suitably broaden a field of view. The image processing apparatus includes: first input means for receiving input of sinogram information acquired by projecting radiation onto an object; means for configuring a first tomographic image of the object from the sinogram information; second input means for receiving input of a prior tomographic image obtained by imaging the object before the sinogram information; conversion means for converting a pixel value of the prior tomographic image on the basis of a pixel value of the first tomographic image; and means for generating a second tomographic image from the sinogram information with use of the prior tomographic image, the pixel value of which has been converted.

The invention discloses a sensitive speaker monitoring and recording control method and system based on deep learning. The method comprises the following steps of: acquiring recorded audio information, extracting an audio characteristic value of the audio information by adopting a triple loss method based on a time pool and cosine similarity, inputting the audio characteristic value into a pre-trained speaker model of the sensitive speaker, and outputting a matching value of the voice matching degree with the sensitive speaker, comparing the matching value with a preset matching threshold value, if the matching value is smaller than the matching threshold value, determining that the speaker is not a sensitive speaker, and not recording the audio information, otherwise, storing the audio information, and meanwhile, updating the speaker model of the sensitive speaker. The method has the advantages that the accuracy of monitoring the sensitive speaker is improved; and the problem that existing recording does not distinguish speakers or cannot accurately judge a target speaker is solved.

The invention relates to the technical field of material processing, in particular to a high-pressure casting device for a cyclopentane household appliance new material. The casting device comprises abase, a conveying belt for conveying a mold is fixedly installed on the base through a bolt, a clamping device used for clamping and positioning the mold is fixedly installed on the base through a bolt, and a supporting frame is fixedly installed on the base through a bolt. The casting device has the beneficial effects that when the high-pressure casting device is used, raw materials can be directly foamed and mixed through a mixing tank so as to prepare a mixture suitable for casting, and then high-pressure casting can be directly carried out; and a casting system, which is composed of a discharging pump, a first electromagnetic material valve, a material barrel, a second lifting push rod and a second electromagnetic material valve, of the casting device can effectively improve the casting pressure, and can also conduct accurate quantitative casting, so that incomplete finished products caused by insufficient casting pressure and insufficient casting amount are reduced, and the product quality is effectively improved.

The invention refers to providing a system that allows to reduce the computational costs when using an iterative reconstructional algorithm. The system (100) comprises a providing unit (110) for providing CT projection data, a base image generation unit (120) for generating a base image based on the projection data, a modifying unit (130) for generating a modified image, wherein an image value of a voxel of the base image is modified based on the image value of the voxel, and an image reconstruction unit (140) for reconstructing an image using an iterative reconstruction algorithm that uses the modified image as a start image. Since the modifying unit is adapted to modify the base image, the base image can be modified such as to form an optimal start image for the chosen iterative reconstruction such that a faster convergence of the iterative reconstruction can be accomplished.

The invention discloses a planning sub-target merging method based on deep learning, which mainly solves the problem of granularity upper limit selection of sub-target merging in the preprocessing process of a planning system, and comprises the following steps: generating a data set, and performing data expansion and data cleaning on samples; training the obtained sample as a deep learning model,selecting a Resnet deep residual network as the model, and separately training planning problems in different fields to generate a feature model of each field; modifying a preprocessing code of the ASOP planning system, adding an updating function of a sub-target granularity threshold value. Through a trained model, the planning system can determine the upper limit of granularity according to a specific planning problem. According to the method, the upper limit threshold value of the granularity is determined from the macroscopic level through a deep learning method, so that the efficiency andthe quality of the planning system during problem solving are balanced, the planning capability of the planning system is effectively improved, and the labor cost is greatly reduced through dynamic autonomous adjustment of a computer.

The invention discloses a cell image segmentation method based on deep learning, and belongs to the field of image processing. The method specifically comprises the following steps: segmenting cells in test tube holes by using an improved Mask R-CNN (Region-Convolutional Neural Network), obtaining cell images, establishing a digital image data set, fusing feature learning into a model establishing process, and accurately detecting and segmenting the cells in the test tube holes. A quick, low-cost and accurate automatic detection technology is realized, and the detection efficiency can be improved.

The invention discloses a sketch image drawing method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a face image and a corresponding real sketch image; inputting the face image into an encoder of a feature pyramid network for step-by-step encoding to obtain convolution features of different scales; inputting the finally output convolution features into a decoder of the feature pyramid network for step-by-step decoding, and fusing the middle convolution features in the decoding process to obtain a plurality of prediction sketch images; fusing all the predicted sketch images to obtain a final predicted image; calculating image loss of the final prediction image and the real sketch image according to a preset loss function, and optimizing network parameters of the feature pyramid network according to the image loss to obtain an optimal feature pyramid network; and inputting the target face image into the optimal feature pyramid network for sketch image generation to obtain a target sketch image. According to the method, the generated sketch image has richer details and is more natural and clearer.

The invention belongs to the technical field of big data, and provides a data uniqueness verification method and device during batch modification, and the method comprises the steps of building the mapping of index data and pseudo column data in data to be subjected to uniqueness verification, wherein the mapping is a one-to-one correspondence relationship; modifying the pseudo column data according to the batch modification type of the data to be subjected to the uniqueness verification; and checking the uniqueness of the batch modified data according to the mapping and the modified pseudo column data. According to the invention, the defects of the existing data uniqueness verification scheme are overcome, the existing business scene can be basically covered, and the data uniqueness verification requirement is met; and the operation steps of the business personnel during batch operation are reduced, so that the operation time of the business personnel during batch operation is saved.