Abstract: Disclosed is a method for determining a number of product quality incidents based on a convolutional neural network, the method includes the following steps: building an appearance quality dataset, building a quality evaluation model, and analyzing a number of product quality incidents; a preset model and a preset optimized model are trained and validated according to the built appearance quality dataset to obtain initial quality evaluation models, which are filtered to obtain a quality evaluation model according to model performance coefficient and comprehensive model coefficient, product appearance defect data are obtained according to appearance defect data detected and obtained by the quality evaluation model, appearance defect ratio and significant appearance defect ratio of the products under quality inspection are obtained according to the product appearance defect data, and a number of product quality incidents of the products under quality inspection is finally obtained according to an incident occur

Abstract: An authentication device includes a feature amount generation unit for generating a feature amount of an object in an image, a registered data acquisition unit for acquiring registered data where a feature amount of a predetermined object in an image is registered in advance as a registered feature amount, and NG information associated with the registered feature amount unsuitable for authentication is recorded, a similarity acquisition unit for acquiring a similarity between the predetermined feature amount and the registered feature amount acquired from the registered data acquisition unit, and a determination unit for performing authentication if the degree of similarity acquired by the degree of similarity acquisition unit satisfies a predetermined authentication condition, and even if the degree of similarity satisfies the predetermined authentication condition, if the NG information associated with the registered feature amount is acquired, the predetermined feature amount is not authenticated.

Abstract: A circuit device includes an image processing circuit that receives target image data by adding effective image data to known image data, executes image processing on the target image data, and outputs processed effective image data corresponding to a processed effective image and processed known image data corresponding to a processed known image. The processed known image data has a CRC expected value calculated in advance. A CRC calculation circuit obtains a CRC value based on the processed known image data. A determination circuit determines correctness of the image processing by comparing the CRC value with the CRC expected value of the processed known image data. The processed effective image is located in an effective image region that is displayed on a screen. The processed known image is located in a non-display region that is not displayed on the screen.

Abstract: The present disclosure provides a transmission line defect identification method based on a saliency map and a semantic-embedded feature pyramid, including the following steps: step 1: cleaning and classifying a dataset; step 2: generating a super-resolution image for a small target of a transmission line by using an Electric Line-Enhanced Super-Resolution Generative Adversarial Network (EL-ESRGAN) model; step 3: performing image saliency detection on the dataset by constructing a U2-Net; step 4: performing data augmentation on the dataset by using GridMask and random cutout algorithms based on a saliency map, and generating a classified dataset; and step 5: performing image classification on a normal set and a defect set by using a ResNet34 classification algorithm and a deep semantic embedding (DSE)-based feature pyramid classification network.

Abstract: A computer-implemented method, according to various embodiments, includes receiving images such as photographs and videos from one or more portable computing devices associated with one or more individuals (e.g., construction workers or landscapers) while the portable computing devices are in a particular position within a particular location at a particular time. The method further includes determining a virtual position within a representation of the particular location that generally corresponds to the particular position and combining the images with the representation to generate an enhanced representation of the particular location. This may allow, for example, owners of a particular property to track and quickly understand construction and landscaping work that has been done on their property and to easily contact those individuals regarding that work.

Abstract: A computer implemented method (82) for colour video processing, the method comprising: receiving a CFA image from a CFA imaging device, wherein the CFA image comprises CFA image pixels defining a plurality of image regions, each image region having at least one pixel of a first colour and at least one pixel of a second colour, at least one image region representing a portion of an object; receiving a full colour image from a colour imaging device, wherein the full colour image represents the portion of the object; storing the full colour image in a memory; isolating pixels of the first colour of the CFA image to generate a first colour channel comprising pixels of the first colour channel; isolating pixels of the first colour of the full colour image to generate a second colour channel comprising pixels of the second colour channel; processing the pixels of the first colour channel using a first computer vision algorithm to generate first image metadata; processing the pixels of the second colour channel usin

Abstract: This application discloses a video data processing method and apparatus, a computer device, and a storage medium, and belongs to the field of video processing technologies. According to this application, when a second video is embedded into a target sub-image of a first video, a position movement of the target sub-image in the first video is obtained based on an artificial intelligence technology. According to the position movement, motion blur processing is performed on the picture of the second video correspondingly, so that the second video, after being embedded into the first video, will have the same motion blur effect as the target sub-image in the first video. Therefore, the first video and the second video can be better fused, and a better video transition effect can be obtained.

Abstract: A mobile cleaning robot includes a cleaning head configured to clean a floor surface in an environment, and at least one camera having a field of view that extends above the floor surface. The at least one camera is configured to capture images that include portions of the environment above the floor surface. The robot includes a recognition module is configured to recognize objects in the environment based on the images captured by the at least one camera, in which the recognition module is trained at least in part using the images captured by the at least one camera. The robot includes a storage device is configured to store a map of the environment. The robot includes a control module configured to control the mobile cleaning robot to navigate in the environment using the map and operate the cleaning head to perform cleaning tasks taking into account of the objects recognized by the recognition module.

Abstract: An automatic vehicle window cleaning method and apparatus includes obtaining a vehicle window image; determining a dark channel image corresponding to a single frame of vehicle window image; further determining, based on a quantity of pixels whose gray values exceed a preset gray threshold in the dark channel image and/or definition of the dark channel image, that a first-type foreign object exists on a vehicle window, and/or determining, based on RGB values of pixels in i consecutive frames of vehicle window images, that a second-type foreign object exists on the vehicle window, and controlling a cleaning tool to remove a foreign object. The foreign object includes the first-type foreign object and/or the second-type foreign object.

Abstract: A system for intelligent image dehazing based on haze type classification. The system receives a hazy image and classifies the haze type using either a Single Selective Classifier (SSC) or a Hybrid Conditional Classifier (HCC), based on user preference. If SSC is selected, a specific single model is used for classification. If HCC is selected, all pre-trained and pre-defined models are used. The system then selects a suitable specialized dehazer based on the predicted haze type and applies it to the input image to mitigate the effects of the haze. The system is flexible, allowing for adjustments to various parameters as per the users' requirements and target application domains.

Abstract: A learning device performs learning a facial recognition model so as to further reduce a triplet loss that uses a first facial image, a second facial image that is a candidate for an adversarial example of a same person as the first facial image, and a third facial image that is a candidate for an adversarial example of a different person than the first facial image.

Abstract: A neural network is trained for use in a substrate residue classification system by obtaining ground truth residue level measurements of a top layer of a calibration substrate at a plurality of locations, each location at a defined position for a die being fabricated on the substrate. A plurality of color images of the calibration substrate are obtained, each color image corresponding to a region for a die being fabricated on the substrate. A neural network is trained to convert color images of die regions from an in-line substrate imager to residue level measurements for the top layer in the die region.

Abstract: The invention refers to an apparatus for monitoring a subject (121) during an imaging procedure, e.g. CT-imaging. The apparatus (110) comprises a monitoring image providing unit (111) providing a first monitoring image and a second monitoring image acquired at different support positions, a monitoring position providing unit (112) providing a first monitoring position of a region of interest in the first monitoring image, a support position providing unit (113) providing support position data of the support positions, a position map providing unit (114) providing a position map mapping calibration support positions to calibration monitoring positions, and a region of interest position determination unit (115) determining a position of the region of interest in the second monitoring image based on the first monitoring position, the support position data, and the position map. This allows to determine the position of the region of interest accurately and with low computational effort.

Abstract: A processor detects a structure of interest from a plurality of tomographic images indicating a plurality of tomographic planes of an object. The processor selects a tomographic image from the plurality of tomographic images according to a type of the structure of interest in a region in which the structure of interest has been detected and generates a composite two-dimensional image using the selected tomographic image in the region in which the structure of interest has been detected and using a predetermined tomographic image in a region in which the structure of interest has not been detected.

Abstract: A server is provided. The server provides virtual indoor space content corresponding to a real indoor space. The server includes a memory; a communication module; and at least one processor operatively connected to the memory and the communication module. The at least one processor may receive image data to generate the virtual indoor space content from a mobile terminal through the communication module, the image data including, for each of a plurality of indoor points, an omnidirectional image acquired from each indoor point and relative location information on the indoor point, and generate the virtual indoor space content by mapping a plurality of virtual points corresponding to the plurality of indoor points in a virtual indoor space based on the relative location information.

Abstract: A vehicle occupant detection system includes a processor that extracts, using an occupant detection area position information stored in a memory, an occupant detection area within a captured image of an in-vehicle camera, and detects any occupant in that area. The processor executes a correction mode using a correction-purpose image obtained by capturing, using the camera, a linearly-extending metal frame exposed when a vehicle door is open. In the correction mode, the processor detects an actual position of the metal frame in the correction-purpose image, calculates, as a shift amount, a difference between the actual metal frame position and an expected metal frame position in the correction-purpose image, and updates the occupant detection area position information in the memory such that an area reached by moving the occupant detection area of the current captured image by the shift amount serves as the occupant detection area of a new captured image.

Abstract: A method for determining an optimized weighting of an encoder and decoder network; the method comprising: for each of a plurality of test weightings, performing the following steps with the encoder and decoder operating using the test weighting: (a) encoding, using the encoder, a reference image and a distorted image into a latent space to form an encoding; (b) decoding the encoding, using the decoder, to form a distortion map indicative of a difference between the reference image and a distorted image; (c) spatially transforming the distorted image by the distortion map to obtain an aligned image; (d) comparing the aligned image to the reference image to obtain a similarity metric; and (e) determining a loss function which is at least partially defined by the similarity metric; wherein the optimized weighting is determined to be the test weighting which has an optimized loss function.

Abstract: According to an embodiment of the present invention, a multi-resolution image processing device capable of recognition of a plurality of dynamic objects comprises a dynamic object identifier extracting a motion vector from an input image and identifying a dynamic object; a dynamic object ID imparter assigning an ID to the dynamic object identified by the dynamic object identifier; a dynamic object artificial intelligence determiner determining an authenticity of the dynamic object identified by the dynamic object identifier based on a standard shape for each type of dynamic object learned and stored, classifying at least one object of interest and object of no interest designated by a user, and removing the object of no interest; a region-of-interest detector extracting a feature point from the at least one object of interest and setting at least a portion of the object of interest as a region of interest; and a variable compressor compressing the image with different compression rates for the region of inter

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a model to detect objects. One of the methods includes maintaining first training data that represents at least a plurality of first images of invertebrate fossils; determining, using at least one or more geographic properties of a geographic region, one or more transformations from a group comprising: a pixel group translation, a pixel group two-dimensional rotation, a changed background, or an object obfuscation; creating, using the one or more transformations and the first training data, second training data that represents at least a plurality of second images of invertebrate fossils; training, using the second training data, a machine learning model to detect invertebrate fossils in images of the geographic region; and providing, to a system, the trained machine learning model to enable the system to detect invertebrate fossils in images of the geographic region.

Abstract: A sensor and method for sleep position detection including: a transmitter configured to transmit electromagnetic waves between 30 GHz and 300 GHz; a receiver configured to receive the electromagnetic waves from the transmitter, wherein the transmitter and receiver are positioned in relation to person sleeping such that the receiver receives reflected electromagnetic waves; and a control station configured to analyze the transmitted and received electromagnetic waves to determine a position of the person sleeping. In some cases, the method may include: forming a radar cube of results; performing a fast fourier transform (FFT) on the radar cube; applying a constant false alarm rate (CFAR) processor to the FFT data; determining a capon gradient; forming a 5-dimensional feature space based on the capon gradient; and conducting an optimization of SVM.