13results about How to "Increase the learning rate" patented technology

The invention relates to the field of anti-unmanned aerial vehicle systems, in particular to a vehicle-mounted anti-unmanned aerial vehicle intelligent photoelectric device and system. The technical problems are as follows: a miniaturized light-weight photoelectric system adopts an FPGA + DSP photoelectric system to track and detect a small unmanned aerial vehicle, the tracking and identification capabilities are weak, the tracking of the unmanned aerial vehicle is unstable, and a target cannot be detected and identified; according to the technical scheme, the vehicle-mounted anti-unmanned aerial vehicle intelligent photoelectric device and system comprise a photoelectric load assembly and a servo control assembly; compared with the prior art that a miniaturized and lightweight photoelectric system adopts an FPGA + DSP photoelectric system to track and detect a small unmanned aerial vehicle, the tracking and recognition capability is weak, the tracking of the unmanned aerial vehicle is unstable, and a target cannot be detected and recognized, the system adopts an intelligent target detection and recognition model based on a YOLOv5 algorithm, and the system has a wide application prospect in a domestic Hisilicon processor Hi3559AV100. 30 frames of images can be processed per second, the detection precision is high, the result is accurate, the anti-interference capability is strong, and the all-weather and all-time target identification requirement can be met.

The application discloses a new energy station electric quantity prediction method and system based on a global-local double-layer prediction architecture, and aims at solving problems of insufficient monthly electric quantity prediction data of new energy stations, large medium and long term weather prediction errors, and low prediction precision of traditional experience methods, adopts a global-local double-layer model architecture, a XGBoost global prediction model is constructed by using all station data in the upper layer, common rules such as seasons, weather, installed capacity, shutdown plan and the like are mined, the global model output is taken as an enhanced feature, a XGBoost local model is trained in combination with single station data in the lower layer, and individual characteristics such as station equipment efficiency and operation and maintenance level are adapted, and a holiday adjustment factor is introduced to post-process and correct the prediction result. The application effectively solves the small sample modeling problem commonly existing in monthly electric quantity prediction, and significantly improves the monthly electric quantity prediction accuracy of wind power stations and photovoltaic stations.

The application provides a communication interference strategy generation method based on reinforcement learning, comprising the following steps: step one, constructing a system model of a communication party and a communication interference party; step two, based on the system model of the communication party and the communication interference party constructed in step one, learning an anti-interference scheme of the communication party by using a win-or-learn policy hill climbing algorithm, and designing a corresponding interference decision model; step three, using the interference decision model obtained in step two, learning an anti-interference strategy of a communication target and implementing interference according to a decision process of 'observation-adjustment-decision-action'. The application comprehensively considers interference basic principles and behavior changes of the communication target, combines a jam-to-noise ratio and anti-interference behaviors such as frequency change and transmission power increase of the communication target after being interfered, and uses the two as measurement indexes of interference effects, so that the purpose of real-time and rapid interference is achieved.

The invention relates to the technical field of civil engineering, and discloses a slope safety coefficient calculation method based on transfer learning and GRU, and the method comprises the following steps: S1, exploring a slope with a safety coefficient to be calculated, and obtaining the physical and mechanical parameters of the slope; s2, building and training a safety coefficient calculation model; s3, calculating a slope safety coefficient; and S4, carrying out transfer learning. According to the method, various physical and mechanical parameters of the side slope are used as input layers, and corresponding safety coefficients are output through two one-dimensional convolution layers, two GRU layers and two full-connection layers, so that the acquisition efficiency of the safety coefficients is improved, the parameters of each side slope and the safety coefficient of each side slope are used as one sample, and the sampling efficiency is improved. And training a slope safety coefficient calculation model through the sample, reducing the initial learning rate of the source domain model, slowing down the learning rate of the source domain model, increasing the learning rate of a full connection layer, performing multiple tests, and evaluating the accuracy of the model on target domain data.

The target detection model training method based on SE-FPN, the target detection method and the device, the training method comprising: scaling a plurality of training images according to different scaling factors, and splicing them into a new image, wherein the new image comprises a plurality of targets of different sizes; allocating the plurality of targets of different sizes to different pyramid feature layers of the SE-FPN target detection model according to a predetermined allocation strategy; in each pyramid feature layer, finding the target closest to the center point according to the true value of the training sample of the layer m positions, calculate m All anchors at positions and the true value DIoUD g ,calculate D g Mean m g , Standard Deviation v g , get the threshold t g , select greater than t g And the center position and anchor output in the target box; calculate the classification loss function and position regression function, and train the model through the back propagation algorithm. Construct a target detection network model based on SE-FPN, improve the image preprocessing method and sample selection strategy, and train the model, which is applied to target detection to improve the efficiency of target detection.

The invention discloses an automatic monitoring method for a radioactive tailings pond based on multi-source heterogeneous data fusion, which belongs to the field of data monitoring technology, and includes: setting monitoring equipment for a radioactive tailings pond, collecting monitoring data of the radioactive tailings pond, and transmitting all monitoring data to a monitoring platform based on a transmission mode; the monitoring platform constructs a fusion model, and the fusion model fuses all monitoring data based on data type and time characteristics to generate a heterogeneous data combination; constructs a learning model, inputs the heterogeneous data combination into the learning model for learning and training, generates a prediction and early warning model, and extracts state characteristics corresponding to the radioactive tailings pond; outputs a state prediction result of the radioactive tailings pond within a prediction time based on the state characteristics, obtains the accuracy of the prediction and early warning model, and completes the early warning monitoring of the radioactive tailings pond based on the state prediction result, the accuracy and the preset safety threshold. The present invention can improve the timeliness and accuracy of monitoring of radioactive tailings ponds.

This disclosure relates to batch normalization layers. The present invention provides methods, systems, and apparatus for processing input using a neural network system including a batch normalization layer, comprising a computer program encoded on a computer storage medium. One method includes: receiving a corresponding first layer output for each training example in the batch; calculating a plurality of normalization statistics for the batch based on the first layer outputs; normalizing each component of each first layer output using the normalization statistics to generate a corresponding normalized layer output for each training example in the batch; generating a corresponding batch normalized layer output for each training example from the normalized layer outputs; and providing the batch normalized layer outputs as input to a second neural network layer.

The invention discloses an intelligent kit for early screening of urinary system cancer and liver cancer, and belongs to the technical field of embedded medical supplies for early detection of cancers. The intelligent kit mainly comprises biological rapid detection test paper, a sensor, a neural network embedded chip, a microprocessor, a display screen and the like, a blood or urine sample of a patient is dropped on the detection test paper, and after calculation of the embedded chip, a detection result is reflected on external equipment in real time, so that the detection accuracy is improved. And whether the patient suffers from urinary system cancer, liver cancer or other diseases is further distinguished. In this way, early detection of the urinary system cancer and the liver cancer is achieved, the detection efficiency of the urinary system cancer and the liver cancer is improved, comprehensive improvement of functions and performance of portable instruments and large-scale detection equipment is achieved, and great significance is achieved for improving medical conditions of people and improving the living standard of people.

The present application relates to a kind of temperature and pressure integrated composite sensor fast self-adaptive decoupling method, belong to intelligent sensor field, solve the low speed slow problem of temperature pressure decoupling precision.Utilize acquisition model to obtain bridge voltage signal and corresponding temperature value, pressure value, respectively based on bridge voltage signal and corresponding temperature value, pressure value constructs first, second sample data set;Respectively establish temperature, pressure prediction BP neural network model;First, second sample data set is trained to temperature, pressure prediction BP neural network model;Loss function is calculated when each training iteration, and the learning rate of weight and bias is optimized using loss function, and the optimized learning rate is used in the process of reverse propagation to adjust weight bias;When the prediction accuracy meets the requirement training ends, respectively obtain trained temperature, pressure prediction BP neural network model as decoupling model;Real-time bridge voltage signal is input into decoupling model, and temperature pressure is decoupled in real time.Temperature pressure accurate real-time decoupling is realized.

The invention discloses a rule-based scene library setting automatic driving decision-making system and method and a vehicle, and belongs to the technical field of automatic driving, and the system comprises a sensor module which is used for collecting environment information around the vehicle; the data processing module is connected with the sensor module, and the data processing module is used for processing and analyzing the environment information collected by the sensor module so as to identify an obstacle and extract feature information of the obstacle; the decision module is used for making an avoidance decision according to the feature information extracted by the data processing module on the basis of a predetermined rule; and the execution module is used for controlling the vehicle to execute the avoidance decision generated by the decision module. According to the invention, various obstacles and moving states thereof can be accurately identified in a complex urban traffic scene, sudden occurrence of the obstacles can be rapidly judged and processed, and decision rules are continuously optimized through real-time learning and experience accumulation so as to adapt to different urban environments and obstacle arrangement.

The invention relates to an unmanned aerial vehicle control method and system based on wind driven generator blade damage detection, and belongs to the technical field of unmanned aerial vehicles. The method comprises the steps that the real-time distance between the current position of an unmanned aerial vehicle and the center of a wind driven generator is obtained through an RGB-D camera, and the position of the unmanned aerial vehicle is corrected; a wind driven generator image is obtained through an RGB-D camera, a space coordinate system is established, and a wind driven generator blade recognition result is obtained through calculation of an image detection and recognition model; presetting a detection point sequence, obtaining related parameters, and calculating to obtain change coordinates of the detection points through a coordinate calculation model according to the related parameters; inputting the related parameters into an unmanned aerial vehicle control network model to obtain a motor change control quantity of the unmanned aerial vehicle; the intelligent terminal generates a self-adaptive signal according to the motor change control quantity to control the unmanned aerial vehicle to move to a detection point and perform defect detection; and the execution is repeated until all the blades are detected. High-precision unmanned aerial vehicle control is achieved for wind driven generator blade detection.

The present invention discloses a standard text classification method based on machine learning, which belongs to the field of data recognition technology, including adopting the BERT algorithm based on constrained attention to vectorize the standard text; adopting the generative adversarial network algorithm based on semantic focus to generate samples, thereby realizing the expansion of the standard text data after vectorization; and adopting the neural network algorithm based on perturbation fuzzy fusion to extract features and classify the standard text data after vectorization. The present invention uses an interactive attention mechanism and a dynamic context perception mechanism, so that the model can not only capture the semantic association within the text, but also better understand the relationship between the standard text and the classification task instructions, significantly improving the classification model's ability to understand the context. A generative adversarial network based on semantic focus is used to generate additional sample data, and the semantic quality and consistency of the generated data are ensured through a semantic refocusing mechanism, solving the problem of insufficient samples in the standard text classification task.

The present invention provides methods, systems and apparatus for processing inputs using a neural network system including a batch normalization layer, including a computer program encoded on a computer storage medium. One method in the method includes: receiving a corresponding first layer output for each training example in the batch; calculating a plurality of normalization statistics for the batch based on the first layer output; normalizing each component of each first layer output using the normalization statistics to generate a corresponding normalized layer output for each training example in the batch; generating a corresponding batch normalization layer output for each training example in the training examples from the normalized layer output; and providing the batch normalization layer output as an input to the second neural network layer.