49results about How to "Not easy to fit" patented technology

The invention belongs to the technical field of road traffic operation state detection and control, and relates to a road congestion state detection method based on computer vision. The road congestion state detection method specifically comprises the technological steps of firstly classifying and marking data sets formed by processing a large number of collected images to serve as training sets for neural network training, then constructing a convolutional neural network according to the data sets by using a migration model training method, then classifying intercepted real-time traffic monitoring video images through the convolutional neural network, judging the vehicle density state, and finally calculating the optical flow field by using an optical flow algorithm so as to judge the traffic congestion state. The detection method is scientific in design principle and accurate in information collection, the image recognition accuracy reaches 98% and above, the monitoring effect is good, the cost is low, the effect is good, the data calculation method is simple, the judgment accuracy is high, the application is convenient, and the real-time traffic state can be effectively judged.

The invention discloses a target tracking method based on triple twin hash network learning, and relates to the technical field of computer vision, target tracking and deep learning. According to themethod, firstly, a triple twin hash network is constructed, and the network is composed of a data input layer, a convolution feature extraction layer and a hash coding layer. In the initial training process of the network, a training data set and a random gradient descent back propagation algorithm are used for training the triple twinning Hash network, and after training is completed, the initialcapacity of target positioning can be obtained through the network. In the tracking process, firstly, an input image passes through a triple twin region recommendation network to obtain correspondingcandidate frames, then the candidate frames are input into a triple twin Hash network to be subjected to forward processing, the similarity between each candidate frame and a query sample is calculated, the candidate frame with the highest similarity is selected as a tracking target object, and therefore target tracking is achieved.

The invention discloses a transformer fault diagnosis method based on M-RVM fusion dynamic weighted AdaBoost. The method comprises the following steps: a M-RVM classification model is established based on the transformer characteristic data training, then each test sample is tested and the information entropy of each sample is calculated, the training samples are screened by information entropy, and the selected samples are used to train the base classifier A-MRVM based on AdaBoost, finally, the samples to be tested are classified and classified by M-RVM classifier and the information entropyis calculated, and the information entropy is compared with the information entropy threshold, if the information entropy is smaller than the threshold, the classification result of M- RVM classifieris as the output, whereas multiple A-MRVM base classifier are used to continues to classify the samples, based on the classification of the tested samples of each A-MRVM base classifier, the weightedcoefficient of the base classifier and the final strong classifier is weighted and integrated to improve the diagnostic accuracy of the whole algorithm.

The invention belongs to the technical field of image processing, provides a three-dimensional CT image segmentation method based on a multi-view separation convolutional neural network, and mainly relates to three-dimensional automatic segmentation of a skeleton in the CT image by using a novel convolutional neural network. The method aims to solve the problems that a neural network using three-dimensional convolution is too large in model, too high in running memory occupation amount and incapable of running on a small-video-memory-capacity display card or embedded device. Meanwhile, in order to improve the capability of the convolutional neural network for utilizing the three-dimensional space context information, a multi-view separation convolution module is introduced, the context information is extracted from the multi-view sub-images of a three-dimensional image by using a plurality of two-dimensional convolution, and the multi-level fusion is carried out, so that the extractionand fusion of the multi-view and the multi-scale context information are realized, and the segmentation precision of the skeleton in the three-dimensional CT image is improved. The average accuracy of the improved network structure is obviously improved, and the number of model parameters is obviously reduced.

The invention discloses an unbalanced data classification method based on mixed sampling and machine learning. The method comprises the steps of step 1, generating a training set; step 2, for a few types of sample sets P in the training set, copying P to generate P ', using P and P' to synthesize PP ', adopting an smote algorithm to generate S on the basis of the PP', and P, P 'and S form PP' S at the same time; step 3, for the majority of types of sample sets N in the training set, randomly undersampling without putting back to obtain t Ni; step 4, repeatedly executing the step 2 for t timesto obtain t different PP 'Si, and synthesizing Ni and the corresponding PP' Si into a new training set to obtain t subsets; step 5, training to generate t classifiers Hi; and step 6, integrating t Hito obtain a final classifier H, and utilizing the classifier H to complete classification of the unbalanced data set. According to the method, the attention of few types of samples is improved, and meanwhile information of multiple types cannot be excessively lost; The possibility of over-fitting and over-generalization is reduced; The training effect is good, overfitting is not prone to occurring, and the training speed is high.

The invention discloses a method for efficiently and rapidly regenerating adventitious buds from pear leaves. The method comprises steps as follows: acquiring sterile tissue culture seedlings, generating adventitious buds by inducing the leaves and performing primary culture and subculture on the adventitious buds, wherein an inducing medium is a combination of a basic culture medium NN69, auxin IBA and cytokinin TDZ, the leaf age is about 20-50 d, dark culture is performed for 21 d, the regeneration effect of leaves of Pyrus bretschneideri Rehd. is remarkable, the regeneration rate can reach 70.83%, the average number of regeneration buds of each leaf is 2.06, the callus occurrence rate reaches 100%, the browning rate is very low and is even 0, and demands for germplasm preservation of pear varieties and requirements for the regeneration rate of the Pyrus bretschneideri Rehd. serving as a genetic transformation material are met. Requirements for the leaves are low, the material sources are enriched, and the problem of high browning rate in the woody plant tissue regeneration process is solved. The problem that the regeneration rate of the oriental pear varieties is generally low is effectively solved, an effective material is expected to be provided for genetic transformation of the oriental pear varieties, and breakthrough of genetic transformation of pears is realized.

The invention relates to the field of change detection of computer vision, in particular to an image change detection method integrating a residual error network and a UNet network, which comprises the following steps of: transforming a coding part of the UNet network into the residual error network, and keeping a decoding part unchanged; generating a twin network from the transformed UNet network, and respectively extracting abstract features of the images in different periods; calculating the difference between the network output and the reference image by using a comparison loss function, and training the modified UNet network; using the trained network to calculate difference graphs of images in different periods, and using an image segmentation technology to search an optimal segmentation threshold to extract a change area; according to the method, the residual network is introduced, so that the problem of gradient disappearance in the layer-by-layer mapping process is avoided; and the UNet twin network is adopted, so that fewer parameters and less data are required during training, and over-fitting is not easy to occur.

The invention discloses an industrial part defect detection method based on deep learning, and relates to the field of industrial quality inspection, and the method mainly comprises the steps: obtaining a preset number of industrial part original images and defect marking graphs after defect marking; obtaining a feature map after convolution pooling processing according to the defect labeling map, fusing the feature map with the output of each pooling layer in the pooling stage, and obtaining a segmentation network by using the initial convolution kernel; adjusting the size of the convolution kernel to train the segmentation network in sequence; performing classification training on output results of the corresponding segmentation networks according to the original images and the defect labeling graphs to obtain classification networks; and judging the defect degree, defect position and defect type of the original image of the industrial part according to the segmentation network and the classification network. According to the method, the problem of defect segmentation is converted into the problem of classification through sequential pooling-up-sampling-fusion processing, and the advantage that the convolutional neural network is good at classification is utilized, so that the high efficiency of defect marking and classification of industrial parts is realized.

The invention discloses an attention mechanism-based lightweight semantic segmentation model construction method, which is applied to the technical field of image processing, and a training set is formed by giving an image I and a corresponding real label graph GT. The method comprises the steps of step 1, establishing a model; step 2, model training; and step 3, model testing: inputting a test set image into the trained network model to obtain a test result. According to the invention, the image segmentation accuracy and segmentation speed are improved; the segmentation process is not easy to over-fit; efficiency is high, and actual deployment is facilitated; and under the condition that the annotation data is insufficient, the annotation data is quickly trained, so that the performance is further improved.

The invention discloses a method and a system for detecting genuine nature of ginseng based on a terahertz time-domain spectroscopy technology. The method comprises the following steps: acquiring a to-be-detected ginseng sample; measuring the thickness, the terahertz time domain signal and the terahertz spectrum of the ginseng sample, and calculating the extinction coefficient of the sample; inputting the extinction coefficient into a trained ginseng origin prediction model, and outputting origin information of the ginseng sample to be detected. The method has the advantages that the strong penetrability and fingerprint spectrum characteristics of terahertz are utilized, the terahertz time-domain spectroscopy technology is combined with machine learning, a supervised multi-classification prediction model is established, model hyper-parameters are optimized through an optimization algorithm, the ginseng producing area prediction model which is high in accuracy, good in generalization performance, not prone to over-fitting and good in anti-noise performance is established.

The embodiment of the invention discloses a model training method and device, a target detection method and device, equipment and a storage medium. The method comprises the steps of acquiring a first sample with an instance-level label and a second sample with an image-level label; the second sample is obtained based on the instance-level label of the first sample; determining a pseudo label of sample data in the second sample through a pre-trained target detection model; determining the original detection loss of the target detection model based on the instance-level label of the sample data in the first sample; determining classification enhancement loss of the target detection model based on a pseudo label of sample data in the second sample; and training the target detection model by using the first sample and the second sample based on the original detection loss and the classification enhancement loss.

The invention relates to a short-term load prediction method based on multi-granularity characteristics and an XGBoost model. The short-term load prediction method comprises the following steps: acquiring historical short-term load data of a to-be-predicted regional power system; analyzing fluctuation influence factors of the historical short-term load data to obtain date granularity information and meteorological granularity information; calculating the correlation between the multi-dimensional granularity of the date granularity information and the meteorological granularity information and the short-term load by using the Pearson correlation coefficient; selecting a feature combination with high correlation according to the correlation; and predicting the short-term load of the screened feature combinations with high correlation through an XGboost model. According to the method, the Pearson correlation coefficient is used for selecting the characteristics with high multi-granularity correlation as the input, the complexity of the model is reduced, and the XGBoost is used as the prediction model, so that the problem of large-scale data classification can be solved, and the method has the advantages of high accuracy, low possibility of over-fitting and high expandability.

The invention discloses a deep learning-based sea wave height prediction method and application thereof, and the method comprises the steps: respectively inputting sea wave data information into an AM-LSTM model and a CatBoost model, obtaining and outputting P1 and P2, and reconstructing P1 and P2 according to the following formula to obtain a prediction sequence P; p = q2 * P1 + q1 * P2, w1 is the mean value of the MAE, the RMSE and the MAPE output by the AM-LSTM model, and w2 is the mean value of the MAE, the RMSE and the MAPE output by the CatBoost model. According to the sea wave height prediction method, the advantages of the LSTM in deep learning in the aspect of processing long-term data prediction, the characteristics of the attention mechanism and the characteristics of few parameters, fast training and difficult overfitting of the CatBoost are considered, the predicted data are reconstructed, the prediction precision is high, the generalization performance is high, the method is especially suitable for sea wave height prediction, and the application prospect is promising.

The invention belongs to the field of machine learning, and particularly relates to a DRBM method based on meta-learning, which divides an algorithm into two stages of meta-learning and model learning by improving a training-testing algorithm of a network. In the meta-learning stage, a training task is utilized to update network parameters, and the updated network parameters are used as initial values of network parameters in the model learning stage, so that the initial values of the network parameters can enable a loss function of network training to descend more quickly and achieve global optimum more easily; and network parameters are updated and tested by using the test task in the model learning stage. According to the algorithm, a meta-learning method is introduced to improve the training process of the DRBM, so that the gradient descent direction of a meta-learning stage of network parameters is descent towards a 'most adaptive' point, and the network can quickly adapt to a new task.

The utility model discloses a soft material sticking structure device. The soft material sticking structure device comprises a power component, a sticking component and a locking component, wherein the power component can open and close the sticking component and vacuumize an inner cabin and an outer cabin of the sticking component; the sticking component is mainly provided with an upper cover and a lower shell; a flexible sticking sheet and an upper outer cabin are arranged in the upper cover; a flexible lower sticking sheet and a lower outer cabin are arranged in the lower shell; the inner cabin is formed between the upper sticking sheet and the lower sticking sheet; the upper cover and the lower shell are buckled through the locking component, so that the upper sticking sheet and the lower sticking sheet can be tightly stuck to two flexible articles to be stuck when preset negative pressure is reached in the inner cabin, and the two articles can be tightly stuck within preset time. The surfaces of the flexible articles are not easily damaged in a processing process, and deformation is not easy to cause, so that the yield of finished products is high and the soft material sticking structure device has industrial utilization value.

The invention relates to an intelligent tooth brushing process evaluation method, which comprises the following steps that: 1) tracking a tooth brushing process of a standard Bass tooth brushing method, and collecting a denoising signal output by a GY-521 gyroscope in an electric toothbrush; 2) checking a data interval preprocessed and labeled in S1), and if the interval time of data is shorter than t seconds, according to the time interval of t seconds, solving a mean value in the period of time; 3) training an evaluation model based on a random forest; and 4) preprocessing the data of a person who brushes teeth according to a sensor data processing method in S1), and using a trained random forest model to classify the toothbrush state of the person who brushes teeth at different moments.The method has the beneficial effects that a machine learning method is adopted to classify the tooth state of the toothbrush in each time slice in a user tooth brushing process, the tooth brushing state of the user at different moments in the tooth brushing process can be predicted, and the tooth cleaning state and strength of the user at different moments can be accurately judged.

The invention discloses a named entity recognition method for Vietnamese, and the method is characterized in that the method comprises the following steps: 1) model training; and 2) data dictionary construction. The model training comprises 1-1) data input, 1-2) BERT layer training, 1-3) GRU layer training and 1-4) CRF layer training, and the data dictionary construction comprises 2-1) data dictionary correction and 2-2) result verified. The named entity recognition method for Vietnamese is high in Vietnamese named entity recognition accuracy.

The invention relates to a control system for a medical transfer equipment, in particular to an air control system used for a negative-pressure isolation transfer device. The system includes a sealingcomponent and a controller which are arranged on a sealing piece, the sealing component includes two half chains arranged oppositely, a magnetic block is fixed on an installation bulge, pressure sensors are fixed on the working face of the magnetic block, and the signal output ends of the pressure sensors are connected with the signal input end of the controller, the output end of the controlleris in signal connection with a negative-pressure pump, the controller receives pressure values sent by all the pressure sensors, all the pressure values are added to obtain the sum of the pressure values, and the controller is used for controlling the reduction of the output power of the negative-pressure pump when the sum of the pressure values is increased. In order to solve the problem that airin an isolation bin is likely to leak due to the arrangement of a sealing zipper, the control system for reducing the possibility of leakage of air in the isolation bin is provided.

The invention relates to a power load prediction method based on a periodic automatic encoder, and the method comprises the steps: collecting the power load data of each transformer area, and forming a historical power time sequence; preprocessing the historical power time sequence, reconstructing the historical power time sequence through the trained periodic automatic encoder, and generating an embedded sequence of the historical power time sequence; and inputting the embedded sequence of the historical power time sequence into the trained predictor to obtain predicted power load sequence data. Compared with the prior art, the method has the advantages of high prediction precision and the like.

The invention discloses a medium and long term load prediction method, and the method comprises the steps: finding out an economic factor which has a long term equilibrium relation with electric quantity through a co-integration test in measurement economics based on a Granger causality test and an LSTM; determining economic factors beneficial to electric quantity prediction by using a Granger causality test method so as to reduce the number of input variables of the prediction model; and finally, inputting the economic factor data into the LSTM model for load prediction. A method of combining the Granger causal relationship test and the LSTM time sequence prediction model is introduced into a multivariable system, the medium-and-long-term load prediction model which is not easy to overfit and high in expandability is constructed, and the model is used for medium-and-long-term load prediction. And the predicted medium-and-long-term load has relatively high accuracy.

The invention relates to a blowing and sucking machine, and belongs to the field of gardening tools. The blowing and sucking machine comprises a blowing pipe, and a roller assembly is installed on theblowing pipe; the roller assembly comprises a roller frame, two rollers and a mounting frame, the mounting frame is fixed to the blowing pipe, the two rollers are arranged on the two sides of the roller frame respectively, a rolling shaft for mounting the rollers is arranged on the roller frame, the rollers are arranged on the rolling shaft in a sleeving mode, the rolling shaft is further provided with a limiting sleeve for limiting falling of the rollers, and the roller frame is slidably mounted on the mounting frame; a plurality of gear protruding blocks are arranged on the mounting frame at intervals in the axis direction of the blowing pipe, limiting protruding blocks are arranged on the roller frame and abut against the gear protruding blocks in the axis direction of the blowing pipe, a deformation notch is formed in the side wall, facing the gear protruding blocks, of the roller frame, and a deformation piece is installed in the deformation notch and fixed to one groove wall ofthe deformation notch; and the limiting protruding blocks are installed on the side wall, facing the gear protruding blocks, of the deformation piece. The machine has the effect that the positions ofthe rollers can be adjusted.

The invention discloses a production process of a foaming super-soft elastic fabric. The production process comprises the following steps that a radial base material is placed on a production device,and a tension standard different from that of a weft base material is set; the weft base material made from two different raw materials is put on the production device, the two raw materials are alternatively arranged, tension stretching is conducted on the raw materials while arrangement, and it is ensured that different raw materials are different in tension; the radial base material and the weft base material are mutually perpendicularly interwoven together to form a woven fabric; high-temperature setting treatment is conducted on the woven fabric; the fabric after the high-temperature treatment is fully cooled; the cooled fabric base material automatically shrinks to form regularly arranged and irregularly-shaped bubbles. Fabrics with uniformly and densely distributed irregular bubbleson the surfaces can be produced by adopting the production process, and ironing operation can be omitted due to the unique appearance of products. Therefore, the fabric is convenient to use and goodin practicability and is not likely to stick to the skin during wearing, and the use comfort of a wearer in summer can be remarkably improved.

The invention provides a loss optimization-based accounting voucher generation method and device and a storage medium, and the method comprises the steps: an optimization step: training an initialized prediction model based on historical data through employing an optimized loss function, and obtaining a trained optimized prediction model; an identification step: using the optimization prediction model to predict and classify the journal data to obtain the category of the journal data; a selection step: selecting a processing strategy corresponding to the predicted category of the journal data from a preset strategy correspondence table; and a processing step of processing the journal data based on the selected processing strategy to generate an accounting voucher. A traditional loss function is optimized, variance calculation is carried out on the first k samples in the n samples according to the sequence from small to large, then calculation is carried out on the last n-k samples with large variance values, and the accuracy of overall certificate generation is improved.

The invention discloses a preparation method of an automobile dash sound insulation pad, wherein the preparation method comprises the steps: S1, material cutting: firstly selecting a cotton felt, a PU foam board and a sound insulation cotton board with proper sizes, then preparing a mold plate of an automobile dash board, putting the cotton felt, the PU foam board and the sound insulation cotton board on the mold plate, and cutting according to the shape of the mold plate. The invention relates to the technical field of sound insulation of automobile dash boards. In conclusion, according to the preparation method of the automobile dash sound insulation pad, through the working procedures from S1 to S4, the cotton felt and the sound insulation cotton plate are bonded through an adhesive, silicon-coated protective paper and a adhesive sticker, so that the cotton felt can be easily taken down, the disassembly difficulty is greatly reduced, the disassembly efficiency is improved, and the replacement cost is low due to the fact that the cost of the cotton felt is low; and finally, a side edge of a final cut edge is wrapped with an adhesive tape, the cut edge is prevented from being polluted, the side edge is wrapped and limited, and the situation that the use quality of the whole connection is affected due to damage of the side edge is avoided.