186results about How to "Learn accurately" patented technology

The invention discloses a target detection method and a target detection system of visual radar spatial and temporal information fusion. The target detection system comprises an acquisition unit, a sampling unit, a superposition unit, a model building unit and an execution unit. The acquisition unit is used for collecting RGB image data and 3D point cloud data to calculate discretized LIDAR depthmap in grayscale; the sampling unit is used for up-sampling and densifying the LIDAR depth map so that the RGB image and the data form of the LIDAR depth map are unified and corresponded to each otherone by one; the superposition unit is used for combining the RGB image and the LIDAR depth map into an RGB-LIDAR picture and superposing the RGB-LIDAR pictures which are continuously collected for multiple times to obtain a superposed RGB-LIDAR picture, wherein the number of the continuous collection of the RGB-LIDAR pictures is equal to or more than 1; the model building unit is used for establishing an RGB-LIDAR data set for the multiple superposed RGB-LIDAR pictures to enter the deep learning network for training and learning and establish a classification model; the execution unit is usedfor taking corresponding decisions according to target analysis results from the classification model. Consequently, the effects of long-distance recognition and high classification accuracy are achieved.

The invention discloses a recommendation method based on deep learning, belongs to the technical field of data mining, and solves the problems of an existing recommendation method that the potential factor vector of a project can not be predicted from text content information which contains the project descriptions and metadata so as to cause recommendation inaccuracy. The method comprises the following steps that: carrying out modeling on the implicit feedback characteristics of the historical behavior data of a user, and learning to obtain the implicit factor vectors of the user and the project after modeling; taking the implicit factor vector of the project as tag training to carry out modeling on the time sequence information of project text contents and deeply mine a network model; and for a new project which does not appear in the historical behavior data of the user, predicting the network model obtained through S(2) in the text content information of the project to obtain the implicit factor vector of the project, directly matching the implicit factor vector of the project with the implicit factor vector, which is obtained in S(1), of the user, and sorting matching degreesto obtain the new project recommendation list of each user. The method is used for recommending new projects.

The embodiment of the invention discloses a human face living body detection system and a method based on optical pulses, which belongs to the field of image recognition. The system comprises a light emitting device, a photographing device and a detection device, wherein the light emitting device emits optical signals of different signal strengths to a to-be-detected object with preset frequency; the photographing device is used for acquiring multiple images generated by the to-be-detected object under the optical signals of different signal strengths; and the detection device is used for detecting whether the to-be-detected object is a living body based on the multiple images. The scheme provided by the embodiment of the invention is applied to target region recognition, and the phenomenon that a person uses non-living objects such as a photo or video to pretend to be the person to pass the target region recognition can be avoided.

The present invention discloses a method and system for controlling an underwater robot locus based on deep reinforcement learning. The system comprises a learning phase and an application phase. In the learning phase, a simulator simulates the running process of an underwater robot and collects the data of the simulated running underwater robot, the data comprises the state of each moment and the target state of the next moment corresponding to each moment, and the learning is performed aiming at a decision neural network, an auxiliary decision neural network, an evaluation neural network and an auxiliary evaluation neural network through the data. In the application phase, the state o the underwater robot at the current moment and the target state of the underwater robot at the next moment are obtained and input to the decision neural network obtained through the final learning in the learning phase, and the decision neural network is configured to calculate the propulsive force required by the underwater robot at the current moment. The method and system for controlling underwater robot locus based on the deep reinforcement learning can realize accurate control of the underwater motion track.

The invention provides a recognition method, based on deep belief network, of three-dimensional SAR images. The method comprises the following steps: firstly establishing a simulation sample bank of the three-dimensional SAR images, performing projection to different azimuthal angles and pitch angles through one or a small quantity of objective three-dimensional SAR images, so as to obtain a plurality of two-dimensional SAR images, ensuring that the small quantity of obtained three-dimensional SAR images are converted into two-dimensional images, and performing recognition through a two-dimensional image recognition method, and the method can greatly reduce the cost, and reduce the time for acquiring SAR imaging. According to the method, a splicing crossover verification method is proposed, and the deep belief network is improved, so that the deep belief network can automatically adjust parameters, self optimization of parameters is realized, the occurrence of over-fitting learning state and under-fitting learning state is effectively avoided, advance features of sample data can be accurately learnt, a better recognition result is obtained for the deep belief network, the complexity of manual setting of parameters is eliminated, and the recognition efficiency is improved.

The invention discloses a construction method and device of a project recommendation model based on a hybrid neural network and a project recommendation method. The construction method comprises the following steps: filtering comment information, preprocessing the filtered comment information, and learning context features related to a project in the preprocessed comment information and user features and project features in scoring information by using a convolutional neural network; subsequently, fusing and interacting the project characteristics in the user-project scoring information and the context characteristics in the comment information, integrating the learned user characteristics and the fused project characteristics into a multi-task learning framework, and performing joint training to obtain a project recommendation model based on the hybrid neural network. According to the invention, the two heterogeneous data of the scoring information and the comment information are integrated into one unified model, so that the implicit feature vectors of the user and the project can be learned more accurately, and the purposes of improving the performance of the recommendation system and improving the recommendation effect are achieved.

The invention provides a recommendation model training method and a prediction method and device based on a recommendation model, and the method comprises the steps: obtaining at least one piece of sample data, and obtaining at least one feature subset according to a feature set of each piece of sample data; determining a condition corresponding to each feature subset in the condition set according to the attribute of each feature subset; wherein the condition set comprises at least two conditions, the at least two conditions respectively indicate different attributes of the feature subsets, and the attribute of each feature subset is consistent with the attribute indicated by the condition corresponding to each feature subset; and training a recommendation model corresponding to a condition corresponding to each feature subset in the model set by using each feature subset and the tag corresponding to each feature subset. By implementing the embodiment of the invention, a better recommendation model can be trained, and the prediction accuracy of the recommendation content is improved.

The invention relates to a forecasting method of burning and explosion characteristics of organic compounds based on a support vector machine; the method utilizes various structural parameters which reflect the structural characteristics of molecules to describe the burning and explosion characteristics of organic matters according to the principle that various burning and explosion characteristics are decided by the molecular structure; the introduction of the support vector machine of a powerful machine learning algorithm can carry out training and forecasting to the burning and explosion characteristics of the organic matter and the nonlinearity, nondeterminacy and complexity which exist in the molecular structure, thus establishing a forecasting module with stability and effectiveness; the established forecasting module is applied in the forecasting of the burning and explosion characteristics of other unknown compounds with the advantages of high forecasting precision, fastness and convenience.

Correction coefficients corresponding to parameters by which characteristic variation is caused are stored in a data memory of a second integrated circuit element in which driving open/close elements respectively connected in series to a plurality of inductive loads are integrated, and a first integrated circuit element that cooperates with the second integrated circuit element suppresses a current control error accompanying individual variation among circuit components and environmental temperature variation by reading present values of the parameters and combining the present values with the correction coefficients. The correction coefficients are calculated by an adjustment tool in a state where the second integrated circuit element is provided singly, and therefore correction data can be generated in relation to a plurality of temperature environments easily.

The invention relates to a three-dimensional target detection method and system based on point cloud weighted channel characteristics, and the method comprises the steps: carrying out the extraction of a target in a two-dimensional image through a pre-trained deep convolutional neural network, and obtaining a plurality of target objects; based on each target object, determining a point cloud viewcone in the corresponding three-dimensional point cloud space; segmenting the point cloud in the view cone based on a segmentation network of the point cloud to obtain an interested point cloud; And based on the network with the weighted channel characteristics, performing 3D Box parameter estimation on the point cloud of interest to obtain 3D Box parameters, and performing three-dimensional target detection. According to the method, the characteristics of the image can be learned more accurately through the deep convolutional neural network. Based on a network with weighted channel characteristics, 3D Box parameter estimation is carried out on the point cloud of interest, the weight of characteristic drop of unimportant points can be reduced, the weight of key points can be increased, interference points can be restrained, the key points can be enhanced, and therefore the precision of 3D Box parameters can be improved.

The invention relates to a commodity recommendation method based on antagonistic network, an electronic device and a storage medium, comprising: obtaining behavior data of a user to a commodity, wherein the behavior data comprises purchase data and browse data; obtaining the behavior data of the user to the commodity; obtaining the behavior data of the user to the commodity. Generating a random input vector according to the behavior data; Inputting a random input vector and behavior data into a countermeasure network model, the countermeasure network model comprising a generator and a discriminator, wherein the random input vector input generator obtains a randomly generated vector, the randomly generated vector is used as a virtual input of the discriminator, and the behavior data is usedas a real input of the discriminator; judging the true input and the virtual input by The discriminator, judging the ratio of the true input and the virtual input by output, and judging whether thediscriminator converges or not. When the discriminator does not converge, the generator is driven to update until the discriminator converges, and the randomly generated vector of the generator is used as a recommendation sequence. The method, apparatus and medium accurately learn the characteristics of the recommending entity.

A control method for a drive apparatus of a hybrid vehicle in which an assist power source is connected to an output member connected to an engine through a torque transmitting member whose torque capacity is changed according to an engagement control amount includes the steps of maintaining a rotational speed of the assist power source at a predetermined rotational speed (step (S02); continuosly changing the engagement control amount while maintaining the rotational speed of the assist power source at the predetermined rotational speed (step S04); and leaning a relationship between output torque of the assist power source for maintaining the rotational speed of the assist power source and the engagement control amount when the output torque of the assist power source reaches a predetermined value while the engagement control amount is changed (step S06).

The invention discloses an unsupervised pedestrian re-identification method based on pseudo-label self-correction, and the method comprises the steps of constructing a source domain data set, a targetdomain data set and a target domain test set, constructing an algorithm model M, carrying out the pre-training of the algorithm model M through employing the source domain data set, and extracting afirst target feature of the target domain data set through employing the algorithm model M; fusing the first target features to obtain second target features, clustering the second target features toobtain pseudo tags, evaluating the quality of the pseudo tags, correcting clusters with poor quality, and taking an obtained result as a pseudo tag repeated training algorithm model M; and extractinga second target feature from the target domain test set by using the algorithm model M and carrying out image matching to obtain a pedestrian re-identification result.

A method for removing rain in a video based on noise modeling is disclosed. Under the assumption of a low-rank background, the rain bar noise component and the moving foreground in the video are simultaneously estimated. First, video data containing rain noise is acquired and a model is initialized; a rain map generation model is created according to the characteristics of the rain noise and the video foreground; the structural characteristics of the rain imaging in the video-a rain bar formed by moving rain droplets on each small block in an image is identical in the direction, the small block prior distribution of the rain bar is established; a moving object detection model is established according to the characteristics of the video foreground sparsity; the model is converted into a rain removal model under the maximum likelihood estimation framework; a rain-containing video and the rain removal model are applied to get a rain-removed video and other statistical variables, and the rain-removed video is output. The method aims to build a high-quality video rain removal model based on a rain map generation principle and rain bar noise structure characteristics, thereby more accurately allowing the video rain removal technology to be widely applied to complex raining scenes with the moving foreground.

The invention discloses a children intelligent conversation learning method and system and electronic equipment. The method comprises the following steps of: guiding a user through a preset guiding voice to adjust an intelligent learning machine, so as to ensure that a camera on the intelligent learning machine is oriented to a to-be-learnt target object; obtaining an image through the camera, analyzing the image and judging whether the image accords with a target object image requirement; if the judging result is negative, continuously guiding the user through the guiding voice to adjust theintelligent learning machine; if the judging result is positive, recognizing a target object image so as to obtain a target object type; obtaining a corpus corresponding to the target object accordingto the target object type; playing a multimedia file, corresponding to the target object, in the corpus and carrying out interactive conversation learning with the user on the basis of the corpus. According to the children intelligent conversation learning method and system and the electronic equipment, interest points and attentions of children users can be correctly grasped, the correctness ofrecognizing voice conversations with the children users is improved, the interaction and communication with the children users can be strengthened and the conversation learning quality is improved.

The invention relates to the technical field of automobile electronics, and provides an engine fuel correction control method. The method comprises the following steps of S1, detecting whether the engine currently meets a fuel correction enabling condition; S2, if the detection result is yes, entering the fuel correction, determining a correction amount and a correction rate of the fuel injectionquantity based on an oil injection closed-loop adjustment coefficient, and correcting the initial fuel value under the current working condition of the engine through the correction of the fuel injection quantity; and S3, when exiting fuel correction, updating the fuel value correction value at the exit moment to the initial fuel value under the current working condition. The control method has the following benefical effects that the latest fuel value learned under each working condition is stored in a self-learning unit corresponding to the working condition, when the engine runs to a similar working condition next time, the last stored fuel correction value can be used as a self-learning initial fuel value, the air-fuel ratio can be adjusted to the vicinity of a theoretical air-fuel ratio more quickly on the basis of last self-learning, and air-fuel ratio self-learning control can be completed through the above process.

An information recording device is used for recording record information in an information recording medium at least including a first recording layer capable of forming a first recording area and a second recording layer capable of forming a second recording area. The information recording device includes: write elements; acquisition elements for acquiring offset information indicating a relative shift; calculation elements for calculating an address (“Y′=Inv Y−α”) indicating a second boundary point opposing to a first boundary point according to the offset information; and control elements for controlling the write elements so as to write record information (i) while making the first boundary point a recording end or start position and (ii) making the second boundary point indicated by the calculated address a recording start or end position.