148results about How to "Avoid training" patented technology

A method of constructing data classifiers in cases where the number of features is much larger than the number of training examples and it is critical to avoid overtraining, is particularly useful in recognizing computer objects containing or likely to contain a computer virus. The method provides a novel way of tolerating imperfectly classifiable data, of learning in time polynomial in the size of the input data, and of avoiding overtraining. Training is performed by constructing and solving a linear program (LP) or integer program (IP).

The present invention belongs to node self-localization field in large-scale wireless sensor network, and discloses a node indoor localization method based on wireless signal intensity, which main includes: getting relationship curve of signal intensity and distance, and putting forward concept of effective wireless signal intensity (Effective RSS, ERSS); obtaining coordinates of target node by selecting localization arithmetic (ERSS filter + maximum likelihood estimation, triangle localization, approximate location estimation) according with anchor point amount through distance between anchor point and target by training RSS curve; providing alerting service by using physics coordinate, providing navigation service according with route arithmetic and address chain by using sign coordinate (for example in hall). Precisely localization of indoor target can be realized in condition of low-cost, simple initial configuration, and a plurality of services based on location can also be provided.

An exercise apparatus includes a motion mechanism and a handle assembly. The motion mechanism is provided for a user to perform exercise. The handle assembly has at least one grip portion for being held by the user during exercising. Specifically, the grip portion has a temperature variable area thereon. The temperature of the temperature variable area could be controlled to rise to a suitable temperature that the hand of the user can feel warm, for example, a temperature between 38 and 50 degrees Celsius. The temperature of the temperature variable area could also be controlled to reduce to a suitable temperature that the hand of the user can feel cool, for example, a temperature between 20 and 36 degrees Celsius. The temperature of the temperature variable area could be directly controlled by the user or automatically changed according to the exercise status of the user.

The invention discloses a product packaging surface defect detection and classification method based on machine vision. The method comprises the steps of 1, acquiring a high-definition color image of defect-free product packaging, making the high-definition color image into a standard image, conducting real-time shooting with a camera, conducting online acquisition of a high-definition color image of product packaging to be detected, and making the high-definition color image as an image to be detected; 2, conducting image matching on the image to be detected and the standard image based on SURF algorithm; 3, conducting difference image operation on the two images matched in step 2 to obtain a defective image; 4, conducting feature extraction on the defective image to obtain the geometrical features and color features of the defective image; 5, classifying product packaging surface defects by means of RBF neural network algorithm. Automatic defect detection and classification are conducted by means of a machine vision system, human factor interference can be avoided, labor cost is reduced greatly, and then huge hidden cost caused by training and management when artificial detection is adopted is avoided.

The invention discloses an unsupervised domain adaptation method based on an adversarial learning loss function, and the method comprises the steps: (1), generating a high-level feature of a source domain image through a feature extraction network G, carrying out the cross entropy loss with a real label through a classifier C, generating a confusion matrix through a domain discriminator D, and correcting a pseudo label into the real label; and (2) generating high-level features of the target domain image through a feature extraction network G, generating pseudo tags through a classifier C, generating a confusion matrix of the high-level features through a domain discriminator D, and correcting the pseudo tags to be in opposite distribution. (3) confronting and optimizing the loss functionby a feature generator and a discriminator. In addition, for the confusion matrix on the target domain, a correction label is generated and serves as a label of the target domain, and the classifier is optimized. By utilizing the method and the device, the noise of the pseudo tag can be corrected in unsupervised domain adaptation, and the distribution difference between the domains is matched, sothat the classification precision of the target domain is improved.

The invention discloses a biofeedback training system which comprises a biological information feedback device and an intelligent terminal. The biological information feedback device at least comprises sensors used for collecting biological information representing a body state; the intelligent terminal comprises a signal receiving unit, an evaluation and training feedback unit and a display unit; the signal receiving unit receives the biological information representing the body state in real time; the evaluation and training feedback unit generating evaluation feedback information according to the biological information received within a period of time, and the evaluation feedback information is used for showing a synchronous trend of the heart rate and respiratory frequency within the period of time; the display unit at least continuously supplies the evaluation feedback information to a user and guides the user to change the biological information by adjusting the body state till the heart rate and the respiratory frequency reach resonance. The biofeedback training system can obtain the biological information and the evaluation feedback information in real time and effectively conduct real-time training guidance on the user, and the training effectiveness and a good using feeling of the user can be improved to a certain extent.

The invention discloses a transverse federation and longitudinal federation combination method and device, equipment and a medium. The transverse federation and longitudinal federation combination method comprises the steps of obtaining available public information, inputting the available public information into a preset longitudinal federation service party; obtaining vector information, and training a longitudinal federation model of the preset longitudinal federation service party based on the vector information, updating the network weight of each preset reinforcement learning model, periodically inputting each updated preset reinforcement learning model into a preset transverse federation server, and iteratively updating each updated preset reinforcement learning model. The technicalproblem of high resource consumption of a computing system of a reinforcement learning model in the prior art is solved.

Method and apparatus for automated operations, such as pruning, harvesting, spraying and/or maintenance, on plants, and particularly plants with foliage having features on many length scales or a wide spectrum of length scales, such as female flower buds of the marijuana plant. The invention utilizes a convolutional neural network for image segmentation classification and/or the determination of features. The foliage is imaged stereoscopically to produce a three-dimensional surface image, a first neural network determines regions to be operated on, and a second neural network determines how an operation tool operates on the foliage. For pruning of resinous foliage the cutting tool is heated or cooled to avoid having the resins make the cutting tool inoperable.

The invention discloses a color image color meaning classification method of based on a fully convolutional network. The method includes 1. constructing a fully convolutional network; 2 obtaining a color image data set with pixel level labeling; 3. training the fully convolutional network by means of the color image data set to obtain the feature model for performing pixel level color meaning classification on color images in any size; 4. performing the pixel level color meaning classification on any color image by the feature model to evaluate the classification precision of the feature model; and 5. optimizing the network classification result by a whole connection condition random field method to obtain the color class label of each pixel in the image, and transforming the color class labels into the corresponding color spaces to display the pixel level color meaning classification result according to the mapping relationship between the class labels and the color spaces. The method provided can realize the pixel level color meaning classification of the color images, and effectively improves the classification precision of the color image color meaning in the complex and changeable environment.

An actuator (1) comprises an electrical motor (10) for driving a damper (3) from a rest position to a stalled position. The damper (3) is biased toward the rest position with at least one spring. A control module (2) is configured to supply, upon activation of the actuator (1), a drive current to the electrical motor (10) for driving the damper (3) from the rest position to the stalled position against a bias force produced by the spring. The control module (2) is further configured to reduce the drive current to a lower level for maintaining the damper (3) in the stalled position when a defined duration of time has elapsed since activation of the actuator (1). Damages to the electrical motor (10) and/or gear trains can be avoided without having to provide any circuitry or sensory equipment for measuring the drive current.

The invention relates to an obstacle detection method and device and a storage medium, and belongs to the technical field of computers. The method comprises the steps: obtaining point cloud data collected by a laser detection module in a process that a moving carrier moves on a moving plane; wherein the point cloud data comprises three-dimensional coordinates of sampling points and reflection signal intensity; projecting the sampling points to a two-dimensional plane according to the three-dimensional coordinates by taking a height direction perpendicular to the moving plane as a projection direction to obtain a bird's-eye view; determining pixel information of each pixel point in the aerial view according to the point cloud data, wherein the pixel information of each pixel point comprisesa first pixel value, a second pixel value and a third pixel value; detecting obstacles in the collection range according to pixel information in the aerial view; the problem of low application deployment efficiency can be solved; by processing the point cloud data with huge data volume into the two-dimensional image data, a large amount of point cloud data does not need to be collected to train adeep learning model, and the application deployment efficiency is improved.

The invention discloses a mobile context point cloud simplification algorithm based on a point feature histogram (namely a PFH operator). The algorithm mainly comprises the steps of receiving a point cloud set from a point cloud obtaining device; calculating the PFH operator of each point; calculating the standard deviation of the PFH of each point, and designing simplification rules; calculating out a corresponding relation table between the threshold values and the simplification rates of different standard deviations; according to the requirements to the simplification rate from a user, determining the threshold value; reserving the point sets with standard deviations less than or equal to the threshold value in point cloud data, in point cloud data, and eliminating other point sets. According to the mobile context point cloud simplification algorithm based on the point feature histogram, the simplification rate is high, the realtime performance is great, and adequate geometrical characteristics are reserved for scene reconstruction.

The invention relates to a federated learning method and system based on a block chain and a trusted execution environment, and belongs to the technical field of artificial intelligence machine learning. A block chain and a trusted execution environment technology are combined, and in a task collection stage, a task owner broadcasts and initiates a crowdsourcing model training task in a block chain network. After the task is received, the node meeting the requirement applies to join a participant contract, and a task publisher randomly selects participants meeting the training requirement number from all applicants and issues the task. The selected participant locally trains the model, and meanwhile, in the TEE environment of the selected participant, the correctness proof of model training is generated by comparing whether the Hash values updated by the model are consistent or not. After all the models are trained and updated, the participants send updated models and certificates to an aggregation contract for model aggregation and verification, and corresponding rewards are issued to the participating nodes after verification is passed. High-confidence verification is realized, and the problem that training participants are not credible is solved.

An apparatus that indicates the speed of oncoming trains at train crossings comprises a large digital display attached to a drop gate or alternately on a side gate crossing sign for those crossings without a gate. The digital display is surrounded by a large sign indicating the speed of an oncoming train. A speed detection device aimed in either direction on the track to determine train speed. An interface converts the detected speed into a digital display. It is envisioned that motorists knowing the speed of an oncoming train will be less likely to try to beat the train through the crossing.

A method and apparatus for predicting logistics services, a readable storage medium, the predicting apparatus for logistics services comprising: acquiring first historical data of logistics services;acquiring a shared parameter, the shared parameter being a shared parameter in a second prediction model corresponding to a second historical data of a logistics service, the second historical data ofthe logistics service being earlier than the first historical data of the logistics service; generating a first prediction model using a machine learning algorithm based on the first historical dataof the logistics service and the shared parameters, the first prediction model comprising: task parameters corresponding to the shared parameters and the first historical data of the logistics service; predicting a logistics service based on the first prediction model, and generating a prediction result. Applying the above scheme, the effect and efficiency of logistics business forecasting can betaken into account simultaneously.

The invention discloses a lane changing decision-making method and device; based on the obtained first motion information of a own vehicle and the second motion information of environment vehicles, the driving environment of the own vehicle is divided into different areas; the relative directions of the environmental vehicles and the own vehicle in each area are the same; a target vehicle in eacharea is determined; a lane changing flag bit of each area is in a safe state or an unsafe state is determined based on the a relative distance and a relative speed between the target vehicle in each area and the own vehicle; and a target lane changing decision is made according to the lane changing flag bit of each area. The driving environment of the own vehicle is divided into different areas; whether the target vehicle generating potential threats to normal driving of the own vehicle exists in all the areas or not is determined, such that the final target lane changing decision is made; thewhole lane changing decision making process is simple and clear, and collection and training of a large amount of data are avoided.

An impulse drive for converting inertial and rotational energy into linear motion includes a pair of thrust mass bearing race, with one of the pair of thrust mass bearing race attached to the interior of the top circular bearing ring and another of the pair of thrust mass bearing race attached to the interior of the bottom circular bearing ring, an inner circular bearing ring is coupled to the pair of thrust bass bearing race and have movable arc segments that move into and away from the path of thrust mass that move radially freely within cavities of a rotating circular disk, the rotating circular disk is moved by a axle drive that varies the rotational velocity of the disk, a mass accelerator girdle is positioned exterior the inner circular bearing ring and coupled to the interior of the top and bottom circular bearing rings, with mass accelerators fastened into the mass accelerator girdle and positioned over pistons of the movable arc segments in such a fashion that the amount of force needed to accelerate the trust masses, and the resulting thrust generated by the device is contained and controlled.

The invention relates to a method for monitoring the proper functioning of at least one first and a second component (109, 110) of a vehicle drive train (100). The first and the second component (109, 110) each have an effect on the torque applied to the driveshaft (111) and are controlled for this purpose by a test signal (112, 113). The test signals (112, 113) are predetermined such that the effects of the first and the second component (109, 110) onto the torque applied to the driveshaft (111) cancel each other out if the first and the second component (109, 110) function properly. When the components (109, 110) function properly, the traveling comfort is not disturbed by an unexpected torque increase or torque decrease during the carrying out of this method. Any perturbation of the vehicle drive train (100) is therefore also avoided.

The invention provides a method for reinforcing a soil body by a combination of organic matter and plant urease. The method includes the following steps of firstly, extracting the soybean urease; secondly, mixing the organic matter such as glutinous rice flour, brown sugar or skim milk powder into a solution of soybean urease in proportion; and thirdly, sequentially injecting the organic matter and urease mixed liquid and consolidating fluid into the soil body in certain proportion. According to the method, nucleation point locations are provided for calcium carbonate precipitation by introducing the organic matter into a urease induced calcium carbonate precipitation (EICP) technology so that calcium ions can be concentratedly adsorbed to anion groups provided by the organic mater and large-grained calcium carbonate crystals can aggregate at the nucleation point locations; the growth habit of mineral crystals is affected, the morphology structure of minerals is improved, and the strength of soil body is greatly improved and enhanced; and the activity of the urease is protected, and the calcium carbonate precipitation efficiency under a high concentration is slightly improved. Themethod is efficient, economical, simple in process, friendly to the environment, convenient to popularize and suitable for large-scale treatment.