Abstract: A multi-directionally guided endoscope includes an endoscope body. The endoscope body is connected to a catheter. At least four pulling wires are evenly distributed in the endoscope body. One ends of the pulling wires are connected to an end of the catheter away from the endoscope body. The pulling wires are respectively fixed to pulling mechanisms. The pulling mechanisms are respectively sleeved on guide rods. The guide rods are fixed in the endoscope body. The endoscope body is provided with a plurality of elongated grooves. One ends of the pulling mechanisms respectively extend from the elongated grooves. The multi-directionally guided endoscope is provided with pulling mechanisms that are simple in structure, convenient to operate, and accurate and stable to pull.

Abstract: A self-locking device of an endoscope is provided. The endoscope includes a housing and a toggle assembly. The toggle assembly includes a toggle support, an elastic support and a toggle handle. The toggle handle is sleeved on the toggle support. An inner side of the toggle support is provided with a toggle lever and a friction piece. One end of the toggle lever is hinged to the toggle handle, and the other end of the toggle lever is not connected to the friction piece. The toggle lever has an L-shaped cross section, with a middle bend fixed to the toggle support. The toggle lever is hinged to the toggle support through a first pin. When being stressed, the toggle lever is rotatable around the first pin, so as to push the friction piece to move.

Abstract: A handheld end of an endoscope includes a housing, an insertion end positioning assembly and a self-locking assembly. The self-locking assembly is hinged to the housing. The housing includes an insertion port. The insertion end positioning assembly includes a positioning member, a first positioning nut and a threaded steel tube. The positioning member is inserted into the insertion port. The threaded steel tube is in threaded fit with the first positioning nut. The first positioning nut is fixed in the positioning member. The insertion end positioning assembly further includes a through hole and a fixing member. The fixing member is fit with the through hole to abut against the threaded steel tube. The threaded steel tube is fit with the positioning member to fix an axial position of the positioning member. A second nut is fit with a fixing screw to fix a rotational position of the positioning member.

Abstract: In a described example, a circuit includes a port power path circuit having a power input and a first overload output. An interlock logic circuit has an interlock input, a current limit input and a current limit set output. The current limit input is coupled to the first overload output, and the interlock input is adapted to be coupled to a second overload output of a second port power circuit. The interlock logic circuit is configured to delay providing an output signal to the current limit set output responsive to an overload signal at the current limit input.

Abstract: A system uses a neural network based model to perform scene text recognition. The system achieves high accuracy of prediction of text from scenes based on a neural network architecture that uses double attention mechanism. The neural network based model includes a convolutional neural network component that outputs a set of visual features and an attention extractor neural network component that determines attention scores based on the visual features. The visual features and the attention scores are combined to generate mixed features that are provided as input to a character recognizer component that determines a second attention score and recognizes the characters based on the second attention score. The system trains the neural network based model by adjusting the neural network parameters to minimize a multi-class gradient harmonizing mechanism (GHM) loss. The multi-class GHM loss varies based on a level of difficulty of the sample.

Abstract: A method for generating training models includes generating a preliminary training model based on a group of first images, the first images including different types of objects; processing a group of second images with the preliminary model to generate a probability array for each of the second images, the probability array indicating likelihoods that an object is a particular type of object; generating correlations between the different types of objects based on the probability arrays; generating a plurality of object groups based on the correlations, where each object group includes a plurality of different types of objects that have a relatively low correlation with the other types of objects in the same object group; and for each object group, generating a final training model based on a group of third images, the third images each including an object having an object type corresponding to one of the object types.

Abstract: The present invention relates to a high-strength and high-toughness high-magnesium aluminum alloy and a preparation method thereof. The high-strength and high-toughness high-magnesium aluminum alloy provided by the present invention comprises the following components in percentage by mass: 5.54-6.80% of Mg, 0.50-0.60% of Mn, 0.12-0.16% of Zr, 0.30-0.36% of Er, less than or equal to 0.3% of Si, less than or equal to 0.2% of Fe and the balance of Al. In the present invention, the content of each element is strictly controlled, after a proper amount of Zr and Er rare earth elements in mass percent are added into the aluminum alloy, the mass percent of Mg element in the aluminum alloy is further controlled, so that the Mg element can interact with the rare earth elements efficiently, and the strength and toughness matching of the aluminum alloy with an Al—Mg—Mn—Zr—Er multi-element system can be improved.

Abstract: A method for generating a neural network for detecting one or more objects in images includes generating one or more self-supervised proposal learning losses based on the one or more proposal features and corresponding proposal feature predictions. One or more consistency-based proposal learning losses are generated based on noisy proposal feature predictions and the corresponding proposal predictions without noise. A combined loss is generated using the one or more self-supervised proposal learning losses and one or more consistency-based proposal learning losses. The neural network is updated based on the combined loss.

Abstract: A method for generating training models includes generating a preliminary training model based on a group of first images, the first images including different types of objects; processing a group of second images with the preliminary model to generate a probability array for each of the second images, the probability array indicating likelihoods that an object is a particular type of object; generating correlations between the different types of objects based on the probability arrays; generating a plurality of object groups based on the correlations, where each object group includes a plurality of different types of objects that have a relatively low correlation with the other types of objects in the same object group; and for each object group, generating a final training model based on a group of third images, the third images each including an object having an object type corresponding to one of the object types.

Abstract: Embodiments of the present invention relate to water leakage detection. Data obtained by a smart electricity meter during a predefined time period is collected, where the smart electricity meter is associated with a water-consuming unit. Usages of certain electric appliances in the water-consuming unit during the predefined time period are determined based on the collected data. The electricity consumption of those appliances is correlated to water consumption. The legitimate water consumption of the water-consuming unit during the predefined time period is then determined based on the usages of the electric appliances.

Abstract: The present disclosure is directed to a technique to reduce latency in feedback-based system performance determination. A system receives, from an application developer device, indications of an in-application event and a first input value for an application content delivery profile. The system receives, via an interface from an application developed by an application developer and executed by a computing device remote from the data processing system and different from the application developer device, a ping indicative of an occurrence of the in-application event on the computing device. The system merges data from the ping with internal data determined by the data processing system to generate merged data. The system determines a predicted performance for the in-application event and provides an indication of the predicted performance. The system configures, responsive to the indication of the predicted performance, the application content delivery profile with a second input value.

Abstract: The present disclosure is directed to a technique to reduce latency in feedback-based system performance determination. A system receives, from an application developer device, indications of an in-application event and a first input value for an application content delivery profile. The system receives, via an interface from an application developed by an application developer and executed by a computing device remote from the data processing system and different from the application developer device, a ping indicative of an occurrence of the in-application event on the computing device. The system merges data from the ping with internal data determined by the data processing system to generate merged data. The system determines a predicted performance for the in-application event and provides an indication of the predicted performance. The system configures, responsive to the indication of the predicted performance, the application content delivery profile with a second input value.

Abstract: A lung lobe contour extraction method aiming at DR radiography, which includes the steps: Obtain the representative lung lobe contour template through offline training; initialize the extraction system of contour lobe area in the chest DR radiography; According to dimension of DICOM image, conduct three-layer pyramid decomposition on the images; Use Gabor filter bank to rebuild pending images; Convert residual error of the rebuilt image by the Gabor filter into black and white image; Use Zhan-Suen refinement algorithm to refine the black and white image; Call every template of the offline training and use them as the convolution kernel operator, conduct convolution on the contour imagep; Filter the local optimum convolution value with optimum possibility from the convolution results and combine the evaluations; Generate the lung lobe contour shape in accordance with the upper and lower templates of optimum matching and its position.

Abstract: A DR radiography lung contour extraction method based on fully convolutional network, which includes the steps: Establish the fully convolutional network structure of lung contour segmentation; Conduct off-line training on the weighting parameters of the fully convolutional network; Read DR image and weighting parameters of the fully convolutional network; Input DR image into fully convolutional network and output segmentation results of image through network terminal with network layer-by-layer feedforward. Establish lung contour in accordance with segmentation results.

Abstract: A DR radiography lung contour extraction method based on fully convolutional network, which includes the steps: Establish the fully convolutional network structure of lung contour segmentation; Conduct off-line training on the weighting parameters of the fully convolutional network; Read DR image and weighting parameters of the fully convolutional network; Input DR image into fully convolutional network and output segmentation results of image through network terminal with network layer-by-layer feedforward. Establish lung contour in accordance with segmentation results.

Abstract: A lung lobe contour extraction method aiming at DR radiography, which includes the steps: Obtain the representative lung lobe contour template through offline training; initialize the extraction system of contour lobe area in the chest DR radiography; According to dimension of DICOM image, conduct three-layer pyramid decomposition on the images; Use Gabor filter bank to rebuild pending images; Convert residual error of the rebuilt image by the Gabor filter into black and white image; Use Zhan-Suen refinement algorithm to refine the black and white image; Call every template of the offline training and use them as the convolution kernel operator, conduct convolution on the contour imagep; Filter the local optimum convolution value with optimum possibility from the convolution results and combine the evaluations; Generate the lung lobe contour shape in accordance with the upper and lower templates of optimum matching and its position.

Abstract: The present disclosure is directed to a technique to reduce latency in feedback-based system performance determination. A system receives, from an application developer device, indications of an in-application event and a first input value for an application content delivery profile. The system receives, via an interface from an application developed by an application developer and executed by a computing device remote from the data processing system and different from the application developer device, a ping indicative of an occurrence of the in-application event on the computing device. The system merges data from the ping with internal data determined by the data processing system to generate merged data. The system determines a predicted performance for the in-application event and provides an indication of the predicted performance. The system configures, responsive to the indication of the predicted performance, the application content delivery profile with a second input value.