34results about How to "Achieve generalization" patented technology

An embodiment of the invention discloses a de-weighting method for a short text. The de-weighting method comprises the steps of obtaining text string information of the short text; performing word segmentation on the text string, and obtaining keywords of the text string according to the word segmentation information of the text string; obtaining a text sub string according to a weight corresponding to the keywords, wherein the text sub string comprises keywords with the number of a threshold value; and removing repeating items of the text sub string. According to the technical scheme provided by the embodiment, by obtaining the keywords of the text string, a generalization performance on the original text string is achieved, and the de-weighting generalization capability and efficiency are improved; and meanwhile, the calculated quantity is low, and a de-weighing effect among multiple text strings is realized.

The invention discloses a Jacquard electronic needle block for an integrated warp knitting machine. A stand column is arranged on a base of the needle block. A via hole is formed in a circuit board. The circuit board penetrates through the via hole through a connecting part and is fixed to the stand column. The connecting part serves as a power source cathode of the circuit board. The circuit board is provided with a conductive pressing needle. A through hole is formed in the base of the needle block. The conductive pressing needle penetrates through the through hole to serve as an anode wire of an input power source. A piezoelectric ceramic piece element is bonded to a Jacquard needle. A coral-shaped stand column is arranged on the base of the needle block. A patch plug is arranged on the circuit board. The Jacquard needle is fixedly clamped to the coral-shaped stand column. The conductive part of the tail portion of the Jacquard needle is inserted into the patch plug. A drive and control circuit and an actuator are integrated, line connection of a system is simplified, and generalization of Jacquard guide bars of the warp knitting machine is achieved.

The invention discloses a satisfaction prediction model training method and device, electronic equipment and a storage medium, and relates to the field of artificial intelligence. According to the specific implementation scheme, an intention prediction model is trained according to an online log of the intelligent dialogue system; and based on the trained intention prediction model, a satisfactiondegree prediction model is trained according to the online log, and the structure of the satisfaction degree prediction model includes the structure of the intention prediction model. According to the technical scheme of the invention, the defect that a satisfaction prediction model in the prior art lacks generalization ability of the same intention can be overcome, the satisfaction prediction model is trained through the trained intention prediction model, and the generalization ability of the same kind of intentions can be realized, so that the same kind of intentions have similar satisfaction, and the accuracy of the satisfaction prediction model can be effectively improved.

The invention relates to a no-reference image quality evaluation method and device under meta-learning, computer equipment and a storage medium. The method comprises the steps of obtaining an input training image set; respectively extracting an edge feature map, a local texture feature map and a visual sensitivity distribution schematic diagram of the images in the image set; taking the training image set, the edge feature map, the local texture feature map, the visual sensitivity distribution schematic diagram, the brightness map and the chromaticity diagram as data sets of six quality-related tasks, and learning a quality prior model by utilizing a meta-learning framework; obtaining an input target task image set; and carrying out fine adjustment training on the quality prior model by utilizing the target task image set to obtain a final quality evaluation model. According to the method, a robust quality prior model is learned through a plurality of quality-related tasks by using a meta-learning method, and then a small number of annotation samples of a target quality evaluation task are input for fine adjustment training to obtain a final quality evaluation model, so that generalization can be quickly achieved.

The invention discloses a generalization processing method, device and apparatus and a computer storage medium, and relates to the technical field of artificial intelligence, in particular to a deep learning technology. According to the specific implementation scheme, the method comprises the following steps: determining a candidate query set similar to a requested query in a query library throughat least one of a literal matching mode, a semantic matching mode and a query rewriting mode; determining a generalized query corresponding to the requested query from the candidate query set by utilizing a query matching model obtained by pre-training, wherein the query matching model is obtained by pre-training based on a cross attention model. According to the method, the generalization of therequested query can be realized.

The invention provides a second-hand vehicle valuation method and a valuation system. The method comprises the following steps of firstly, constructing a supervised learning-based linear model for a predicted vehicle model, and carrying out linear regression training by utilizing historical transaction data of the predicted vehicle model; and then the second-hand vehicle price evaluator performs manual adjustment on the abnormal point until the linear model prediction result reaches a preset target. According to the invention, the manual adjustment system is introduced into the training process of the machine learning model, so that the generalization capability of the machine learning model can be effectively enhanced, and the prediction precision of the machine learning model is improved.

The invention discloses a human body pose recognition method in a two-dimensional video image, comprising the following steps: grouping the original video images according to the size of the scale; calculating a sampling image of a specified scale for each group of images, and calculating the Calculate the HOG; use the HOG prediction of a sampled image in each group to calculate the HOG corresponding to other specified scale sampled images in the group; according to the obtained multi-scale HOG, combined with the trained SVM classifier to detect the different scales of the original video image The human body target area; use the trained random forest classifier to classify the pixels of the detected human body target area, and determine the body part area in the human body target area; connect each body part to form a human body contour, and realize human body posture recognition. By applying the method of the present invention, the calculation speed of the multi-scale bottom layer features is accelerated without reducing the detection precision, and the gesture recognition speed and precision are improved.

The invention provides a method for monitoring operation normalization in a concrete unloading process, and equipment. The method comprises: extracting a gray matrix of a color channel of a concrete unloading operation image, performing edge information extraction on the gray matrix to obtain a feature matrix of the color channel, and performing transverse splicing on the feature matrix to obtain a feature vector of the concrete unloading operation image; taking the number of elements of the feature vector as the number of training samples, initializing a behavior classification model, constructing a loss function for the training samples, constructing unconstrained optimization of a weight vector of a multi-layer perceptron neuron according to the loss function, and calculating an optimal solution of the unconstrained optimization by adopting a back propagation algorithm to obtain the behavior classification model; and monitoring the operation process of concrete unloading by adopting the behavior classification model, and determining whether the operation of the concrete unloading process is standard or not. According to the invention, the accuracy of monitoring whether the operation in the concrete unloading process is standard can be improved, and the monitoring cost is reduced.

The invention discloses a crop classification method and system. The classification method includes: constructing historical crop marker samples according to the historical multi-temporal mid-segment images and historical crop phenology data; constructing high-precision small-area current crop marker samples according to the current multi-temporal high-resolution images and current crop phenology data; according to ResNet model and PSPNet model to establish the remote sensing network RSNet model; pre-train the remote sensing network RSNet model according to the historical training samples to establish the pre-trained remote sensing network RSNet model; fine-tune the pre-trained remote sensing network RSNet model according to the current training samples to establish a fine-tuned remote sensing network RSNet model; establishing an optimal pre-trained remote sensing network RSNet model and an optimal fine-tuning remote sensing network RSNet model; classifying the crops according to the optimal pre-training remote sensing network RSNet model and the optimal fine-tuning remote sensing network RSNet model. By adopting the classification method and system provided by the present invention, the workload of training sample selection can be reduced, and the classification efficiency can be improved.

The invention discloses a simulation scene generalization method and device, equipment and a storage medium. The method comprises the steps that at least two associated test parameters in a simulation scene are extracted; expressing association attributes between the at least two associated test parameters; and generating a target generalization scene set according to the association attributes. According to the method and the device, relevance generalization among the test parameters can be realized, so that large-degree-of-freedom scene generalization at an engineer level is supported, and the development efficiency of simulation work is greatly improved.

The invention relates to a RBFNN-based mesoscopic fuel consumption prediction method, which determines the influencing factors of road energy consumption; divides the vehicle driving trajectory into driving segments; calculates the average energy consumption of vehicles in a form segment; and analyzes the average energy consumption distribution of the road section Calculate the average energy consumption of the road section; determine the parameter settings such as road energy consumption factors; use the obtained data set as the training set of the neural network for model learning; input the test data set, and calculate the prediction result of road fuel consumption. With the support of a large amount of energy consumption trajectory data set, the present invention accumulates a large number of observation samples on input parameters such as road section type, vehicle average speed and output parameters of road energy consumption, and grasps the influencing factors of road energy consumption through the training of observation samples The law of the correlation relationship with the average energy consumption of the road, so that the energy consumption of other road sections in the road network with insufficient energy consumption trajectory samples can be predicted, and the generalization of the energy consumption law has been realized. There is a high degree of monitoring granularity. precision.

The invention provides a concrete unloading process operation normative monitoring method and equipment. The method includes: extracting the grayscale matrix of the color channel of the concrete unloading operation image, performing edge information extraction on the grayscale matrix to obtain the feature matrix of the color channel, and horizontally splicing the feature matrix to obtain the feature vector of the concrete unloading operation image ; With the number of elements of the feature vector as the number of training samples, initialize the behavior classification model, build a loss function for the training samples, build the unconstrained optimization of the weight vector of the multilayer perceptron neuron according to the loss function, and adopt the backpropagation algorithm The optimal solution of the unconstrained optimization is calculated to obtain a behavior classification model; the operation process of concrete unloading is monitored by using the behavior classification model to determine whether the operation of the concrete unloading process is standardized. The invention can improve the accuracy of monitoring whether the operation of the concrete unloading process is standardized, and reduce the monitoring cost.

The invention discloses a microgrid short-term load forecasting method based on independent component analysis and support vector machine, including: using Chebyshevv orthogonal polynomials to perform curve fitting on the historical daily load data of the microgrid area, and adding a penalty function for optimization; updating The historical load data is analyzed independently by using the independent component analysis method to obtain the load independent separation source and mixing matrix of the historical load data; for the meteorological data obtained on the historical day and the forecast day, the principal component analysis method is used to obtain the principal component vector, and combined with The date data are used together to form decision-making data; the mixing matrix of historical load data is used as the target data, combined with the historical daily decision-making data, the support vector machine and the fireworks algorithm are used to optimize its parameter penalty factor and kernel parameters, and obtain the optimal support vector machine training model ; Substitute the forecasted daily decision-making data to obtain the forecasted daily load curve. The invention can remove noise interference, and can effectively and accurately predict the future short-term load of the microgrid.

The invention belongs to the technical field of radioactive waste treatment, and relates to a cement solidification treatment method for radioactive waste. The radioactive waste is a radioactive boron-containing concentrate. The method sequentially comprises the following steps: (1) firstly adding the radioactive waste into a waste barrel, then adding lime and an additive, stirring the mixture for10-15 minutes, then adding cement, and continuously stirring the mixture for 10-15 minutes; and (2) capping the stirred waste barrel, and then transferring the waste barrel to a curing room for curing. By utilizing the cement solidification treatment method for the radioactive waste, aiming at the problems existing in the practical application of the American Westinghouse solidification technology, the package capacity of the radioactive boron-containing concentrate is improved on the premise that original solidification equipment and technological operation are kept basically unchanged, andit is ensured that various properties of a waste cement solidification product meet the standard requirements.

The embodiment of the present invention discloses a method and device for automatic driving simulation testing of a scene library. The method includes: determining the traffic test scene based on the target scene template and its corresponding modification instruction, and the modification instruction is to modify the scene element corresponding to the target scene template Instruction; if it is detected that there is a positional relationship between the scene object of the traffic test scene and the vehicle under test, and the scene object that reaches the test trigger condition corresponding to the preset test event, triggers the test process corresponding to the preset test event; monitor the preset The behavior state of the vehicle under test in the test process corresponding to the test event, and obtain the behavior state parameters corresponding to the behavior state of the vehicle under test, so as to determine the test results of the algorithm under test of the vehicle under test through the behavior state parameters, so as to realize the automatic driving scene The automation and comprehensive construction of the automatic driving algorithm can improve the testing efficiency of the automatic driving algorithm while testing the automatic driving algorithm more comprehensively.

The invention discloses a method for conducting arm control through deep learning. The method comprises the steps that firstly, an image of the motion process of a mechanical arm is collected, meanwhile, a control command of the arm is recorded at a certain frequency, and a controller expressed through a deep neural network is obtained through an end-to-end training method. On that basis, close observation shows that the control structure is expressed through the deep neural network, the motion error of the arm can be further reduced through the end-to-end training method, and obstacle avoidance motion can still be achieved greatly under the condition that an obstacle exists. The method is flexible to implement, samples needed by training are decreased to a great extent, and the method has a great advantage for the condition that large samples cannot be obtained easily for motion of the mechanical arm.

The invention discloses a manifold learning generic algorithm based on local linear regression and belongs to the technical field of data dimension reduction of hyperspectral images. The manifold learning generic algorithm based on local linear regression is suitable for any manifold learning algorithms and can keep original manifold learning dimension reduction results. The method includes the first step of searching for a neighborhood, the second step of calculating a projection matrix, the third step of obtaining a linear regression coefficient matrix and the fourth step of calculating a dimension reduction result of a new sample. The generalization of the new sample is achieved on the basis of keeping the original manifold learning dimension reduction results, linear mapping from high dimension to low dimension is established, any manifold learning algorithms such as LE, LLE and LTSA having no generalization ability can be made to have the generalization ability, and therefore the time-consuming manifold learning algorithms are applicable to the dimension reduction process of the hyperspectral images.

The invention provides a portable air conditioner suitable for various occasions, which comprises an indoor unit and an outdoor unit, wherein the indoor unit and the outdoor unit are connected with each other via an indoor and outdoor connecting assembly; fast switching of the indoor and outdoor connecting assembly is realized via a fast connecting component; the indoor unit comprises an automatic drainage system which is characterized in that a water level induction passage is formed on the inner side surface of a clamping groove formed in one end of an indoor unit water-containing plate; an electromagnetic induction switch is in interference fit with an electromagnetic induction switch clamping groove; a floating ball is limited in the water level induction passage; an indoor unit drainage pipe is connected to the water-containing plate; an outdoor unit drainage pipe is connected to an outdoor unit water pump; after the indoor unit drainage pipe and an outdoor unit connecting water pipe are butted via the fast connecting component, an automatic drainage passage is formed; both the electromagnetic induction switch and the water pump communicate with a controller; and the controller controls the on / off of the water pump according to water level information of the electromagnetic induction switch. Due to the automatic drainage system, the portable air conditioner can run continuously to ensure the comfort of a user; and moreover, as manual drainage is avoided, the portable air conditioner is convenient and practical for the user.