32results about How to "Avoid Evaluation Errors" patented technology

The invention discloses a quick and precise high-throughput drug screening system based on deep learning. The drug screening system comprises a picture preprocessing module and a neural network module, wherein the picture preprocessing module comprises a channel merging module and a picture standardization module; the channel merging module merges different cell single-color channel pictures intothe multi-channel picture representation, wherein the merged picture tensors are represented as [H,W,C]; the picture standardization module standardizes the input multi-channel picture data into the tensor representation of [70,70,C]; the neural network module is connected with the picture standardization module, the input data of the neural network module is standardized picture tensors, and finally the trained neural network is subjected to final prediction classification prediction. The drug screening system DeepScreen based on deep learning has the advantages of being high in throughput, precise, efficient, fast, convenient, low in cost and resistant to interference, and has a worth-concerning practical and application prospect.

The invention relates to a wind turbine generator bearing fault diagnosis method for a multi-channel deep convolutional neural network. The method comprises the steps of: simultaneously acquiring high-frequency vibration acceleration signals of a drive end and a non-drive end of a test bearing in various states by using a vibration acceleration sensor; analyzing the acquired vibration signals by using a time frequency analysis technology to obtain corresponding time frequency spectra; establishing a deep convolutional neural network diagnosis model, and training the diagnosis model by using the time frequency spectra and the states of the bearing as a training sample; evaluating the diagnosis model, and applying the diagnosis model to a bearing to be monitored. The method can realize automatic feature learning, avoids feature engineering, effectively utilizes multi-channel vibration signals, and has good universality and extensibility.

The invention provides a method and device for evaluating the influence of a drag torque on vehicle energy consumption. The method comprises a step of measuring a test vehicle equipped with a brake caliper through a chassis dynamometer to obtain first internal resistance, a step of performing an energy consumption test on the test vehicle equipped with the brake caliper to obtain first energy consumption, wherein the motor brake of the test vehicle equipped with a brake caliper can be carried out through an accelerator pedal, a step of carrying out internal resistance measurement on a test vehicle which is not equipped with the brake caliper to obtain second internal resistance through the chassis dynamometer, a step of performing an energy consumption test on the test vehicle which is notequipped with the brake caliper to obtain second energy consumption by using a single pedal gear, wherein the motor brake of the test vehicle which is not equipped with the brake caliper can be carried out through the accelerator pedal, and a step of evaluating the influence of the drag torque on the vehicle energy consumption according to the first internal resistance, the second internal resistance, the first energy consumption and the second energy consumption. In this way, the accuracy of evaluation can be improved.

The invention is applicable to the technical field of aircraft icing detection, and provides an improved method for determining the installation position of the icing sensor: calculating the air flow field of the aircraft with the sensor; calculating the water collection coefficient β1 at the position of the sensor and the The ratio β1 / β2 of the maximum water collection coefficient β2; if the ratio β1 / β2 is greater than 1, calculate the ratio h1 / h2 of the average icing thickness h1 on the sensor to the average icing thickness h2 on the windward side of the wing; if the ratio h1 / h2 is greater than 1, then this position is determined as the installation position of the sensor; if the ratio h1 / h2 is less than or equal to 1, then this position is not suitable for installing the sensor. The invention couples the ice type calculation with the liquid water analysis to judge the installation position of the sensor, avoids the evaluation error caused by the low freezing rate, and has higher evaluation reliability.

The invention discloses a LINUX kernel reliability evaluating system based on source code analysis. The LINUX kernel reliability evaluating system comprises a master control end and an object end, wherein the master control end runs in an independent system, and comprises a control module with control logics, a fault injector module, a communication module, a fault load generator module and a data analysis module; and the object end comprises an injection actuator module, a work load generator module and a result collecting and performance monitoring module. Evaluating errors generated by the traditional method can be avoided effectively, analog internal memory mistakes are injected in a kernel of an object operating system, response of the object system to injected fault is observed, experiment information is collected, the collected experiment information is counted and analyzed, an experiment process of a relevant result is provided for a user, fault injection success rate and fault coverage rate are increased, detection coverage rate and recovery coverage rate of the system are monitored accurately, and accurate evaluation reliability of the system is improved.

The present invention relates to a channel estimator used in a wireless local area network, mainly by using a simplified least square recursion (RLS) channel estimation controller and using a data-reconstruction device, in the transmission of signal packets During the period, the action of adjusting the channel response in the frequency domain is performed, and the adjustment action is flexible and adaptively adjusted at any time point between a transmission signal packet, which can effectively achieve fast convergence and have the best execution channel. Evaluate the effect.

The embodiment of the invention discloses an audio call evaluation method, an evaluation device and a computer storage medium. The evaluation effect of the audio call is applied to the evaluation device, including: obtaining the recognition text of the received audio, wherein the recognition text is: the recognition result of the received audio after the reference audio is transmitted at both ends of the transceiver; combined with the recognition The text and the reference text corresponding to the reference audio are used to obtain an effect evaluation value of the audio call.

The invention discloses an objective evaluation method for quality of VoIP non-reference video communication based on a neural network, is mainly applicable to monitoring the quality of the VoIP video communication under a real-time communication scene, and belongs to the field of objective evaluation of video quality. The method provided by the invention is implemented by the following steps of extracting a network parameter during a VoIP video communication process, meanwhile performing an objective test to score the video, performing standardization treatment and dimension reduction treatment on the extracted network parameter, using acquired dimension-reduced data as an input of the neural network, and using an objective test score as a reference output of the neural network to perform neural network training, thus acquiring a weight and a threshold of each neural node, that is so say building of the neural network is completed, wherein the output of the neural network is the objective quality score acquired through mapping according to the input parameter, and the objective quality score is used for objectively evaluating the quality of the VoIP non-reference video communication. According to the method provided by the invention, the quality of the VoIP video can be accurately evaluated in real time.

The invention discloses a method for evaluating the white blood cell image definition. The method comprises: acquiring an input original white blood cell image and converting the original white bloodcell image into a grayscale image; dividing the grayscale image into image blocks, and randomly extracting the to-be-analyzed image block; performing denoising on the extracted image block to obtain anoise reduction image; performing cell region segmentation of the noise reduction image by using the OTSU method to obtain an optimal segmentation threshold; performing low-pass filtering on the noise reduction image to acquire a degraded image; acquiring TenenGrad parameters of the noise reduction image and the degraded image corresponding to the noise reduction image, and calculating the definition of the image block, wherein the optimal segmentation threshold is used as a customization image threshold of a TenenGrad function; and calculating the mean of the definitions of all extracted image blocks, and analyzing samples to obtain a virtual focus degree threshold. The present invention achieves local evaluation of the image definition while evaluating the overall definition of the image by analyzing definition parameters at different positions of the image.

The invention discloses a fast, accurate and high-throughput drug screening system based on deep learning. The drug screening system includes a picture preprocessing module and a neural network module. The picture preprocessing module includes a channel combining module and a picture standardization module; channel combining The module merges different cell single-color channel pictures into a multi-channel picture representation, and the combined picture tensor is expressed as [H, W, C]; the picture normalization module normalizes the input multi-channel picture data to [70, 70, C ] tensor representation; the neural network module undertakes the picture standardization module, and its input data is a standardized picture tensor, and the final prediction and classification judgment is obtained through the trained neural network. The deep learning-based drug screening system DeepScreen established by the present invention has the advantages of high throughput, precision, high efficiency, fast and convenient, low cost and anti-interference, and has a promising practical application prospect.