139results about How to "Address imbalances" patented technology

The invention relates to a regional-education cloud platform and an education method. The regional-education cloud platform mainly comprises a comprehensive information portal, a teacher-student management space, a cloud platform application center, a regional big data analysis platform, an integration system set, a public cloud resource library and a regional resource converging manager. The whole platform system is of an SOA distributed architecture based on a cloud architecture mode and adopts a B / S development programming mode, and all parts are set or installed on a hardware computer of aplatform so as to form an organic integer and are in electrical connection or software connection. A local server does not need to be established by each school during the regional-education cloud construction, and all data are acquired from a cloud end, so that the cost for deploying local server of the scheme is omitted, the construction efficiency is increased, the expense of information fundis saved, meanwhile, the sharing and common construction of localized characteristic teaching services and resources are provided, and an educational informationalized public service system covering the local region is formed.

The invention discloses a micro-grid multi-inverter parallel voltage unbalanced compensation method. The method relates to an unbalanced compensation ring, a power droop control ring and a voltage and current ring. On the basis of traditional power droop control, three-phase negative sequence voltages and currents are detected, the negative sequence reactive conductance Q-G unbalanced droop control ring is introduced, directive current reference values are synthesized and revised, and then unbalanced compensation of micro-grid voltage is achieved. Through P-f, Q-E and Q-G droop control, distributed power inverters can independently adjust and output fundamental wave frequency, voltage amplitude and unbalanced compensation conductance, and therefore active and reactive equilibrium distribution can be achieved among the inverters. Quasi-resonance PR control is adopted by the voltage and current control ring, astatic control of the voltage is achieved, dead-beat control is adopted, and then accurate control of currents in an inner ring is achieved. According to the method, the three-phase inverters in a micro-grid possess the unbalanced compensation capacity, and therefore micro-grid three-phase voltage balance is maintained.

The invention relates to a photovoltaic module unsupervised defect detection method based on a GAN improved algorithm. The method comprises the following steps: step 1, training a generative adversarial network module according to an SSIM-GAN algorithm model; 2, training an encoder network module according to the SSIM-GAN algorithm model; 3, constructing a defect discrimination module; step 4, carrying out image detection. The method provided by the invention has the beneficial effects that the method can effectively detect tiny and diverse defects of the photovoltaic module, and can solve theproblem of unbalanced samples; an SSIM-GAN model is constructed, and the difference between the images is described by using structural similarity; a normal image is generated through a generative adversarial network; the images are mapped to the corresponding hidden spaces through the encoder network, whether the images to be detected have defects or not is judged through the defect detection module, rapid and accurate detection of unknown defects can be achieved under the condition that the number of the defect images is small, and the method has the advantages of being high in environmental adaptability and robustness.

The invention discloses an image processing method and system, and image processing model training method and system. The image processing method comprises: obtaining a to-be-detected image; inputtingthe to-be-detected image into a neural network model for processing to obtain a bone segmentation result, a bone center line segmentation result and a bone fracture detection result; wherein the neural network model is determined by machine training learning based on a training image. The bone segmentation, bone center line segmentation and bone fracture detection functions are achieved at the same time through the trained deep learning network, the total consumed time can be shortened by 50%, the memory space of the model can be saved by 40%, and meanwhile a doctor can be helped to reduce the film reading burden, accelerate the film reading time, reduce the missed diagnosis probability and reduce the contradiction between the doctor and the patient.

The invention discloses an adenosine triphosphate binding site predicting method for protein. The adenosine triphosphate binding site predicting method for the protein comprises the following steps: firstly, acquiring evolution information and secondary structural information of the protein by using an IPSI-BLAST and PSIPRED program, and extracting characteristics of each amino acid residue by a sliding window technology; secondly, performing random downsampling on non-binding site samples for multiple times by a random downsampling technology, sample of several; thirdly, training an SVM (support vector machine) based on a non-binding site sample subset obtained in each random downsampling and a binding site sample set, and performing random downsampling on all the sample sets to obtain a plurality of SVMs; and finally, integrating the trained SVMs through Dempster-Shafer theoretic evidence. The adenosine triphosphate binding site predicting method for the protein has the advantages as follows: by a random downsampling technology, the scale of a training set can be effectively reduced and the model training speed can be effectively increased; and by an SVM integrating technology, information loss caused by downsampling can be effectively reduced and the model predicting precision can be effectively improved.

The invention discloses a microgrid system and a control method. The microgrid system comprises a plurality of energy storage converters connected in parallel between a DC bus and an AC bus; the AC sides of the energy storage converters are connected in parallel through an AC bus and then connected with the grid-connected or parallel control cabinet; the grid-connected or parallel-connected control cabinet receives an instruction of a coordination control device of the microgrid and controls the microgrid to work in a grid-connected mode or an off-grid mode; the grid-connected or parallel-connected control cabinet obtains a current component reference value of a current inner loop through outer loop control, and sends the obtained current component reference value to each energy storage converter connected in parallel; and each energy storage converter performs current inner loop operation according to the received current component reference value to obtain a driving signal for driving the switching tube of the energy storage converter to be switched on and switched off. According to the method, the information processing speed of the microgrid during information processing can beincreased, Meanwhile, the problems of respective sampling of each energy storage converter in the microgrid energy storage system and imbalance of outer loop calculation errors can be effectively eliminated.

The invention discloses an NIN neural network-based malicious code variation detection method, and belongs to a malicious code protection technology in the field of information safety. The method mainly comprises the following steps of: mapping a malicious code into a grayscale image; enhancing data; and designing a grayscale image detection-oriented NIN neural network. The method specifically comprises the following steps of: firstly, mapping a binary file of the malicious code into the uncompressed grayscale image through a corresponding method; secondly, in order to enable the NIN neural network to better extract features of the grayscale image generated by the malicious code and solving the problem that the image is too small in data scale, previously processing the grayscale image byusing a data enhancement method; and finally, carrying out training by using an NIN neural network model on the basis of existing data.

The invention relates to a digital auxiliary means suitable for the scope of medical treatment and health, in particular to an improvement for health diagnosis and demonstration by using a 3D human model. The invention provides an autonomous human disease diagnosis model, an artificial intelligence physical examination method and system, and a health assessment ocular demonstration method. Based on intelligent equipment, a user autonomously uses the autonomous human disease diagnosis model, the artificial intelligence physical examination method and system, and the health assessment ocular demonstration method, the related disease and the probability value are analyzed and searched through any input physiological data of the human body, then the calculation result is more accurate and reasonable, meanwhile, the various data information and the digital 3D human model are subjected to data connection, the data are loaded on the corresponding part of the model, the various data are highlighted, mapped and demonstrated and are intuitively shown on data processing equipment, so that the difficult problems that the speciality of the medical field is too strong and the content is abstractand not intuitive are solved.

The invention relates to the technical field of fault diagnosis of equipment with high fault rate and discloses a method for diagnosing faults of a wind generating set gear case. The method for diagnosing the faults of the wind generating set gear case comprises the steps that (1) a plurality of effective characteristics of the gear case are selected to establish a multi-dimensional objective space, a plurality of fault categories are simulated, and fault data of the fault categories are collected; (2) time-domain analysis and width-domain analysis are conducted on the fault data and frequency-domain parameters and width-domain parameters are extracted to serve as a training dataset; (3) the training dataset is regarded as a cluster, each datum in the training dataset is regarded as an element, and one data subset is optimized through the iterative algorithm; (4) the speed and the position of each element are updated through a fitness function f(i); (5) an optimal dataset is obtained. The method for diagnosing the faults of the wind generating set gear case solves the problem that data in the wind generating set gear case fault diagnosis are unbalanced and the forecast accuracy of the unbalanced dataset by a classifier is improved by the PSOEE algorithm.

The invention provides a method and a device for replacing data in a caching module. The method comprises the following steps: receiving a new data block or writing information of the new data block into the caching module; selecting a caching block as a preparatory replacing block from the caching module according to the quota in the condition that the whole space of the caching module is full, wherein the caching block is the data block stored in the caching module; and replacing the preparatory replacing block by the new data block. According to the method and the device for replacing the data in the caching module, one caching module is selected as the preparatory replacing block from the caching module according to the quota, so that the quantity of the caching blocks (in the caching module) of a mirror image or a virtual machine can be limited, and the space of the caching module is prevented from being greatly occupied by the highly active mirror image or virtual machine, and as a result, the problem that the caching spaces of the mirror images or virtual machines with different active levels are not balanced can be solved.

The invention provides a mixed reality teaching environment system and an interaction method. The system at least comprises an AR display device. The AR display device provides a virtual environment for displaying different teaching environment related content options covering the real environment according to the space size of the real environment; and according to the image / image of the real teacher, the AR display device provides an image / image used for displaying the virtual teacher, and according to the real teaching aid, the AR display device provides a 3D virtual teaching aid triggeredby the real teaching aid, the user performs teaching interaction with the virtual teacher and the 3D virtual teaching aid by using AR display device. The system has the advantages that the practicalteaching requirements of multi-dimensional interactive experience are met, the system is not limited by time and space, the interest of children in things can be improved in the teaching environment,the children can be attracted to learn and participate in teaching, the bottleneck of live-action practice is effectively broken through, and the problem of unbalanced urban and rural education resources can be solved.

The invention provides a mixed reality remote teaching interaction and teaching aid interaction system and an interaction method thereof. The mixed reality remote teaching interaction system at leastcomprises an AR display device, and the remote teaching interaction system projects matting images of a remote teacher in at least one second teaching scene in a view field of the AR display device ina first teaching scene in real time. The beneficial effects of the invention are that the interaction system breaks through the space limitation of conventional education, so as to achieve the interaction between teachers and students in different spaces.

The invention discloses a network intrusion detection method based on a conditional variation auto-encoder, belongs to the technical field of network space intrusion detection, and the method comprises the steps: firstly training a conditional variation auto-encoder which takes a logarithmic hyperbolic cosine function as a loss function for the expansion of a data set, and then training a classifier based on the convolutional neural network through the data set after data expansion to serve as an intrusion detector of the whole model. According to the method, the classification detection performance on a network space intrusion data set is improved by utilizing the generation capacity of conditional variation self-encoding and the excellent feature extraction capacity of the convolutionalnetwork for data features, and meanwhile, the problem of performance reduction caused by imbalance of data set samples is also solved.

The invention belongs to the field of cloud security technology and virtual desktop, and relates to a resource dynamic scheduling method oriented to virtual desktop cloud, comprising the following steps: S1, establishing an overall system architecture of the resource dynamic scheduling method oriented to virtual desktop cloud; the architecture of the whole system comprises a physical device layer,a pooling layer, a virtual device layer and a resource scheduler. S2, on the basis of the step S1, the resource scheduler realizes the functions of resource allocation, device monitoring and resourcerecovery. The invention realizes automatic monitoring, recovery and redistribution of resources in the virtual desktop cloud environment, and maximally improves the overall utilization rate of resources.

The invention discloses a lymph node detection method based on an improved SegNet segmentation network. The method comprises the following steps: dividing a lymph node image data set into a training set and a test set; Constructing a SegNet segmentation network based on a cavity convolution operation; Training a SegNet segmentation network by using the training set, minimizing a sine and cosine cross entropy loss function as a network optimization target function, and optimizing the SegNet segmentation network; And identifying and segmenting the lymph nodes in the lymph image to be identifiedby using the trained SegNet segmentation network. According to the method, the characteristics are extracted by using hole convolution, the receptive field area is increased under the condition that the additional calculation amount is not increased, the loss of down-sampling information is avoided, and the problem that the resolution of the sampled image is reduced is solved. And through the sineand cosine cross entropy loss function, a weight smaller than that of the cross entropy loss function is given to a sample with a small prediction error, and the problem of unbalanced training of positive and negative samples is solved. And a Markov random field is used to carry out post-processing on the segmentation result, thereby realizing further refinement of the edge part of the segmentation object.

The invention provides a license plate type identification method and device, and the method comprises the steps: carrying out the positioning segmentation of a license plate in a license plate imagethrough a pre-built license plate region detection model; correcting the segmented license plate; according to the corrected license plate, obtaining a license plate type to which the license plate belongs through a pre-established license plate type identification model; wherein the license plate type comprises a preset first type license plate and a preset second type license plate; when the license plate belongs to the first type license plate, taking the license plate type output by the license plate type identification model as a license plate type identification result of the license plate image; when the license plate belongs to a second type of license plate, performing character segmentation on the license plate, and obtaining a character located at a preset position in the license plate as a to-be-identified character; obtaining a character recognition result through a preset character recognition model according to the to-be-recognized character; obtaining a license plate type recognition result of the license plate image according to the character recognition result. According to the method, the accuracy of identifying various license plate types can be improved.

The invention discloses a detection method based on fusion of a simple neural network and an extreme gradient promotion model. The method comprises steps of obtaining a data set, carrying out the preprocessing of the data set, and dividing the data set into a training set and a test set; performing data increment operation on minority class samples in the data set to balance the data set; trainingthe fusion detection model by using the training set; wherein the fusion detection model comprises a simple neural network and an extreme gradient boosting model, and supervised learning training isperformed on the data set till the model converges; and performing intrusion detection on to-be-detected data by utilizing the converged fusion detection model to obtain an intrusion detection result.According to the method, the limitation of insufficient generalization ability of a single machine learning model in different scenes is improved, and the defect of poor association rule mining ability of machine learning for deep information is overcome, compared with a traditional method, cost of manually mining association rules is saved, data features are more effectively utilized, and the intrusion detection rate is increased.

The invention relates to a parallel energy-storage battery state of charge balance control method and apparatus. The balance control method comprises the steps of collecting state of charge (SOC) values of each parallel energy-storage battery in an energy storage power station, and performing sorting on the SOC values based on a shell sort method; detecting whether the sorted SOC values are out of limit or not; and finally allocating charging-discharging power for the energy storage batteries according to a detection result. By virtue of the scheme, the imbalance problem existing the current parallel energy-storage battery SOC can be effectively solved.

The invention belongs to the technical field of image processing, and particularly relates to a cascade detection method of a millimeter wave image human body carried object. In order to solve the problem that a human body carrying object in a millimeter wave image is small, the cascade detection method adopts a Top-down structure to obtain context information of the millimeter wave image, and completes positioning and recognition of a small target through a context relation; in order to solve the problem that a positive sample in a millimeter wave image is sparse, the cascade detection methodadopts a cascade model mode, utilizes a cascade model of the first stage to filter a negative sample while adjusting the coordinate positions of initialized candidate frames of the model so as to provide effective candidate frame information for the cascade model of the second stage; and based on candidate frames with balanced proportion of positive and negative samples and accurate coordinate positions, the cascade model of the second stage further improves the detection rate of the model and reduces the false alarm rate of the model.

The invention relates to an equalization system of a lithium ion battery and an equalization method thereof. The equalization system comprises a battery module, a sampling control module, and an equalization control module, wherein the battery module is formed by 12 battery strings. The sampling control module is used for collecting voltage values of all single batteries in the battery module and a temperature value of the battery module as well as carrying out sampling on currents. According to the collection result, a to-be-equalized single battery group is determined; the to-be-equalized single battery group comprises a single battery in the battery module or comprises two or over two adjacent single batteries. And the equalization control module is used for controlling the to-be-equalized single battery group to carry out charging or discharging. According to the provided equalization method in the invention, even if multi-time charging and discharging are carried out, an occurrence of a non-equalized phenomenon between batteries can be reduced, thereby improving the battery module performance and prolonging the battery service life.

The invention relates to a real-time video face key point detection method based on deep learning, and the method employs a convolutional neural network to carry out the key point detection of a single frame, employs a depth separable convolution to improve the model detection rate, employs a boundary heat map as an additional subtask of an original network to improve the constraint of a global face structure of the original network. The method improves the detection accuracy of an original network, is used for solving a data imbalance loss function of a heat map, improves the generalization capability of a model for a large attitude sample under a limited sample, and improves the inter-frame smoothness through an optical flow loss function. In the detection process, for a frame of which the confidence is lower than a key point confidence threshold due to an extremely large angle, fitting is carried out by utilizing 3DMM to obtain dense key point coordinates, 68-point sampling is carried out on the obtained dense key points according to a projection error between minimum frames, and the consistency with the previous frame is kept. The method has the advantages of real-time performance, capability of utilizing global inter-frame information, high detection accuracy of a face large posture condition and the like.

The invention discloses an image retrieval model training method, an image retrieval method, a storage medium and computer equipment. The training method comprises the steps of obtaining a training sample set, wherein the training sample set comprises a scene picture set shot by a user and a high-definition advertisement scene picture set; Constructing a feature extraction network and an attributeclassifier; Inputting pictures in the training sample set into a feature extraction network to train the feature extraction network, and outputting a training feature vector set; And inputting the feature vector set into an attribute classifier to train the attribute classifier. The retrieval method comprises the following steps: respectively inputting an image to be retrieved and an image in animage library into an image retrieval model, and respectively outputting a feature vector to be retrieved and an image library feature vector set by the image retrieval model; Calculating a Hamming distance value between the to-be-retrieved feature vector subjected to Hash coding and each feature vector in an image library feature vector set; And sorting the images in the image library according to the sequence of the Hamming distance values from small to large.

The invention discloses expanded feed for resisting the fatty liver of tilapia and a preparation method of the expanded feed. The feed is prepared from, by mass, 3% to 8% of domestic fish meal, 15% to 25% of bean pulp, 15% to 30% of double-low rapeseed meal, 10% to 15% of cottonseed meal, 15% to 23% of strong flour, 8% to 15% of corn, 6% to 12% of DDGS, 1% to 2% of freshwater fish oil, 1% to 3% of soybean oil, 1% to 2% of monocalcium phosphate, 1% to 3% of bentonite, 1% to 2% of premix of vitamin and mineral salt, 0.1% to 0.5% of crystalline lysine, 0.1% to 0.5% of free methionine and 1% to 2% of compound fatty liver resisting preparation. Through reasonable component matching, all the nutritional requirements of tilapia are met, the stability of the feed in water is good, the tilapia grows fast, the feed conversion rate is high, and the survival rate is high. Meanwhile, according to the feed, the secretion and metabolic capability of the liver cells of the tilapia can be obviously improved, the liver function and disease resistance of the tilapia can be improved, the organ-body ratio and the fat-body ratio can be decreased, the meat taking rate can be increased, and culture benefits are good.

The invention discloses a lung CT image computer aided system and method based on cluster analysis. The method comprises extracting the nine textural features of a lung CT image including energy, entropy, correlation, difference moment, contrast, sum average, variance, difference, and difference average based on a gray-level co-occurrence matrix; dividing samples into a training set and a verification set according to a ratio of 3:1; subjecting original high-dimensional data to dimensionality reduction by using an improved U-reliefF feature weight calculation method and calculating the corresponding weight values of respective textural features; and applying the weight values to an improved Weightedk-means algorithm to construct a nodule classification model. The method finds, by combininga plurality of textural parameters, that the combination of energy, contrast, entropy and correlation has the best classification effect, achieves a recognition rate of 81.18% of benign nodules and 91.48% of malignant nodules, has a good benign and malignant nodules recognition rate, and contributes to the early diagnosis of the lung cancer.

The invention relates to a defect detection network construction method, an anomaly detection method and system, and a storage medium. The construction method comprises: obtaining a sample image in apreset open data set and a reference image of a standard product of a to-be-detected object; configuring a convolutional neural network module or a plurality of convolutional neural network modules with different scales and forming a defect detection network; training a main feature extraction model by using the sample image to obtain corresponding network parameters, and inputting the reference image into the main feature extraction model and the slave feature extraction model to obtain a corresponding first feature vector and a corresponding second feature vector respectively; and constructing a loss function of the slave feature extraction model according to the first feature vector and the second feature vector, and training and learning to obtain network parameters of the slave feature extraction model, thereby configuring and forming a defect detection network. The defect detection network can complete automatic extraction of features according to input image information, effectively reduces dependence on experience of workers in the process of product defect detection, and has practical value.

An embodiment of the invention discloses a method, equipment and a system for allocating balanced access resources, which belong to the field of time division synchronous code division multiple access systems. The method comprises the following steps that: network-side receives a message which carries the frequency-band capacity of user equipment and is sent by the user equipment; the network-side equipment allocates radio resources to the user equipment according to the message; and the network-side equipment informs the user equipment of radio resource allocation information. The network-side equipment comprises a receiving module, an allocating module and an informing module. The user equipment comprises a sending module, the receiving module, a configuration module and the informing module. The technical scheme of the embodiment of the invention can solve the problem of radio resource unbalancing, thereby being favourable for balancing the allocation of the resources in each frequency band reasonably, reducing the probability of interference and resource congestion and improving network connection rate and user quality experience.

The invention discloses a water surface floating object detection and identification method based on an improved YOLOv3 identification model, and relates to the technical field of computer vision. The method comprises the following steps: pre-collecting water surface floating object data, carrying out the enhancement and amplification of image data through geometric transformation and color transformation, marking the floating objects in the data, obtaining a water surface drifting object data set, and splitting the water surface drifting object data set into a training set and a test set; constructing an improved YOLOv3 network model, and training the improved YOLOv3 network model by adopting the water surface drifting object training set; constructing a water surface drifting object test set according to the water surface drifting object data image, and detecting and identifying the water surface drifting object test set by using the trained improved YOLOv3 network model. The improved YOLOv3 has strong generalization ability, occupies small storage space and video memory space, improves detection and identification accuracy, can ensure real-time performance, and can realize accurate and rapid monitoring and identification of water surface drifts in client equipment with limited computing power and memory.

The invention discloses health-care corn oil containing peony seed oil and a preparing method thereof. The health-care corn oil adopts edible vegetable oil rich in omega-3 and omega-6 polyunsaturated fatty acids as a raw material and is edible oil which is formed by mixing corn oil and the peony seed oil. The health-care corn oil containing the peony seed oil, disclosed by the invention, has the advantages of solving the problem that the proportions of fatty acids in the conventional corn oil are out of balance and in particular, the proportions of the omega-3 and omega-6 fatty acids are out of balance, having a simple formula, enabling the preparation of the corn oil to be more reasonable and being the novel health-care corn oil with more comprehensive nutrition, good mouth feeling and better health-care performance.

The present disclosure relates to a training method and a segmentation method and apparatus for a segmentation learning network of a 3D image and a medium. The 3D image comprises a target object and the proportion occupied by the target object is lower than a preset threshold value. The method comprises the following steps that: a segmentation learning network is constructed based on the sequential combination of a plurality of dense blocks, dense connection exists among basic units in the dense blocks, and each basic unit consists of a batch normalization layer, a RELU layer and a rolling layer; a segmented learning network is trained by a processor based on a training dataset of the 3D image using a loss function that focuses more on difficult samples and penalizes negative voxels that are far away from the object of interest. The method can rapidly and accurately segment irregular and small attention objects by using the learning network with a more compact structure and fewer parameters, and the training process of the learning network can solve the unbalance problem of samples and foreground and background and avoid the problem of over-fitting caused by lack of training samples as much as possible.