Described is a system that can recognize novel objects that the system has never before seen. The system uses a training image set to learn a model that maps visual features from known images to semantic attributes. The learned model is used to map visual features of an unseen input image to semantic attributes. The unseen input image is classified as belonging to an image class with a class label. A device is controlled based on the class label.

The disclosure is directed to an object detection method using a CNN model and an object detection apparatus thereof. In an aspect, the object detection method includes generating a sensor data; processing the sensor data by using a first object detection algorithm to generate a first object detection result; processing the first object detection result by using a plurality of stages of sparse update mapping algorithm to generate a plurality of stages of updated first object detection result; processing a first stage of the stages of updated first object detection result by using a plurality of stages of spatial pooling algorithm between each of stages of sparse update mapping algorithm; executing a plurality of stages of deep convolution layer algorithm to extract a plurality of feature results; and performing a detection prediction based on a last-stage feature result.

A method includes determining, at an image processing device, object quality values for a plurality of objects represented in an image. The object quality values are based on portions of image data for the image. The object quality values include a blurriness value for each object and a color value for each object. The method includes accessing, via the image processing device, object category metrics associated with an object category. The object category metrics include a blurriness metric for each object and a color metric for each object. The method also includes performing, with the image processing device, a particular image processing operation for the image based on comparisons of the object quality values for each object to corresponding object category metrics.

A calculation method includes: receiving a calculation instruction of a parse neural network, obtaining a weight CO*CI*n*m corresponding to the calculation instruction according to the calculation instruction; determining a KERNEL SIZE of the weight, scanning the weight with the KERNEL SIZE as a basic granularity to obtain a weight identifier, storing KERNEL corresponding to a second feature value of the weight identifier, deleting KERNEL corresponding to a first feature value of the weight identifier; scanning all values of the weight identifier; if the value is equal to a second specific value, extracting KERNEL and input data corresponding to the value, performing computation of the input data and the KERNEL to obtain an initial result; if the value is equal to the first feature value, not reading KERNEL and input data corresponding to the value; performing computation of all the initial results to obtain a calculation result of the calculation instruction.

Image data is run through a neural network, and the neural network produces a vector representation of the image data. Random sparse sampling masks are created. The vector representation of the image data is masked with each of the random sparse sampling masks, the masking generating corresponding sparsely sampled vectors. The sparsely sampled vectors are transmitted to nodes of a consensus network, wherein a sparsely sampled vector of the sparsely sampled vectors is transmitted to a node of the consensus network. Votes from the nodes of the consensus network are received. Whether a consensus is achieved in the votes is determined. Responsive to determining that the consensus is achieved, at least one of identification and verification of the image data may be provided.

A method is provided to perform local de-noising of an image. The method includes obtaining a region of interest and a region of noise within a scan. The method also includes determining, for a first image based on the region of interest and a second image based on the region of noise, sample blocks and atoms for each image, where each atom contributes to a weighted sum that approximates a sample block in the image. The method also includes determining a measure of similarity of each atom from the first image with atoms from the second image and removing an atom from the first image if the measure of similarity exceeds a predetermined threshold value. The method also includes reconstructing a de-noised image based on atoms remaining in the first image after removing the atom from the first image, and presenting the de-noised image on a display device.

Image data is run through a neural network, and the neural network produces a vector representation of the image data. Random sparse sampling masks are created. The vector representation of the image data is masked with each of the random sparse sampling masks, the masking generating corresponding sparsely sampled vectors. The sparsely sampled vectors are transmitted to nodes of a consensus network, wherein a sparsely sampled vector of the sparsely sampled vectors is transmitted to a node of the consensus network. Votes from the nodes of the consensus network are received. Whether a consensus is achieved in the votes is determined. Responsive to determining that the consensus is achieved, at least one of identification and verification of the image data may be provided.

An apparatus to facilitate processing of a sparse matrix is disclosed. The apparatus includes a plurality of processing units each comprising one or more processing elements, including logic to read operands, a multiplication unit to multiply two or more operands and a scheduler to identify operands having a zero value and prevent scheduling of the operands having the zero value at the multiplication unit.

A system and method are provided for taxonomically distinguishing grouped segments of signal data captured in unconstrained manner for a plurality of sources. The system comprises a vector unit constructing for each of the grouped signal data segments at least one vector of predetermined form. A sparse decomposition unit selectively executes in at least a training system mode a simultaneous sparse approximation upon a joint corpus of vectors for a plurality of signal segments of distinct sources. The sparse decomposition unit adaptively generates at least one sparse decomposition for each vector with respect to a representative set of decomposition atoms. A discriminant reduction unit executes during the training system mode to derive an optimal combination of atoms from the representative set. A classification unit executes in a classification system mode to discover for an input signal segment a degree of correlation relative to each of the distinct sources.

The present invention discloses a superpixel classification method based on semi-supervised K-SVD and multiscale sparse representation. The method includes carrying out semi-supervised K-SVD dictionary learning on the training samples of a hyperspectral image; using the training samples and the overcomplete dictionary as the input to obtain the multiscale sparse solution of superpixels; and using the obtained sparse representation coefficient matrix and overcomplete dictionary to obtain the result of superpixel classification by residual method and superpixel voting mechanism. The proposing of the present invention is of great significance to solving the problem of salt and pepper noise and the problem of high dimension and small samples in the field of hyperspectral image classification, as well as the problem of how to effectively use space information in classification algorithm based on sparse representation.