In a method for performing adaptive content classification of a video content item, frames of a video content item are analyzed at a sampling rate for a type of content, wherein the sampling rate dictates a frequency at which frames of the video content item are analyzed. Responsive to identifying content within at least one frame indicative of the type of content, the sampling rate of the frames is increased. Responsive to not identifying content within at least one frame indicative of the type of content, the sampling rate of the frames is decreased. It is determined whether the video content item includes the type of content based on the analyzing the frames.

Disclosed are a method, terminal and computer readable storage medium for image classification. The method includes: determining an image feature vector of an image based on a convolutional neural network, where the image comprises textual information; determining a text feature vector based on the textual information and an embedded network; determining an image-text feature vector by joining the image feature vector with the text feature vector; and determining a category of the image based on a result of a deep neural network, where the result is determined based on the image feature vector, the text feature vector and the image-text feature vector.

A method of updating a classifier on-the-fly is provided. The method includes providing a base classifier. The base classifier is a neural network. The method further includes receiving a class and a set of images associated with the class. The method further includes splitting the set of images into an evaluation set and a training set. The method further includes updating the base classifier on-the-fly to provide an updated classifier. Updating the base classifier includes (1) extracting features for each image from the training set from the base classifier; (2) training the updated classifier using the extracted features; and (3) scoring the evaluation set with the updated classifier.

A method for shadow and cloud masking for remote sensing images of an agricultural field using multi-layer perceptrons includes electronically receiving an observed image, performing using at least one processor an image segmentation of the observed image to divide the observed image into a plurality of image segments or superpixels, extracting features for each of the image segments using the at least one processor, and determining by a cloud mask generation module executing on the at least one processor a classification for each of the image segments using the features extracted for each of the image segments, wherein the cloud mask generation module applies a classification model including an ensemble of multilayer perceptrons to generate a cloud mask for the observed image such that each pixel within the observed image has a corresponding classification.

A system for detecting clouds and cloud shadows is described. In one approach, clouds and cloud shadows within a remote sensing image are detected through a three step process. In the first stage a high-precision low-recall classifier is used to identify cloud seed pixels within the image. In the second stage, a low-precision high-recall classifier is used to identify potential cloud pixels within the image. Additionally, in the second stage, the cloud seed pixels are grown into the potential cloud pixels to identify clusters of pixels which have a high likelihood of representing clouds. In the third stage, a geometric technique is used to determine pixels which likely represent shadows cast by the clouds identified in the second stage. The clouds identified in the second stage and the shadows identified in the third stage are then exported as a cloud mask and shadow mask of the remote sensing image.

A system for detecting clouds and cloud shadows is described. In one approach, clouds and cloud shadows within a remote sensing image are detected through a three step process. In the first stage a high-precision low-recall classifier is used to identify cloud seed pixels within the image. In the second stage, a low-precision high-recall classifier is used to identify potential cloud pixels within the image. Additionally, in the second stage, the cloud seed pixels are grown into the potential cloud pixels to identify clusters of pixels which have a high likelihood of representing clouds. In the third stage, a geometric technique is used to determine pixels which likely represent shadows cast by the clouds identified in the second stage. The clouds identified in the second stage and the shadows identified in the third stage are then exported as a cloud mask and shadow mask of the remote sensing image.

A neural network-based classification method, including: obtaining a neural network and a first classifier; inputting input data to the neural network to generate a feature map; cropping the feature map to generate a first cropped part and a second cropped part of the feature map; inputting the first cropped part to the first classifier to generate a first probability vector; inputting the second cropped part to a second classifier to generate a second probability vector, wherein weights of the first classifier are shared with the second classifier; and performing a probability fusion on the first probability vector and the second probability vector to generate an estimated probability vector for determining a class of the input data.

A communications device (100) for classifying an instance (110) using Machine Learning (ML) is provided. The communications device is operative to acquire a feature vector representing the instance, classify the instance using a local first ML model, calculate a confidence level, and, if the calculated confidence level is less than a threshold confidence level, acquire information identifying one or more other communications devices, and transmit a classification request message comprising the feature vector to the one or more other communications devices. The one or more other communications devices are selected based on at least one of: an identity of a user of the communications device, a contact list of the user, a type of data comprised in the feature vector, an origin of the feature vector, the classification of the instance using the local first ML model, a location of the communications device, a respective location of the one or more other communications devices, a location associated with the instance, and one or more classified instances which are related to the instance represented by the feature vector.

Systems and methods relating to multitask transfer learning. Neural networks are used to accomplish a number of tasks and the results of these tasks are used to determine parameters common to these and other tasks. These parameters can then be used to accomplish other related tasks. In the description, data fitting as well as image related tasks are used. Task conditioning as well as the use of a KL regularizer have greatly improved results when testing the methods of the invention.

A hybrid inference facility receives a sequence of data items. For each data item, the facility: forwards the data item to a server; subjects it to a local machine learning model to produce a local inference result for the data item; and the local inference result to a queue; aggregates the inference results contained by the queue to obtain an output inference result; and removes the oldest inference result from the queue. The facility receives from the server cloud inference results each obtained by applying a server machine learning model to one of the data items forwarded to the server. For each received cloud inference result, the facility substitutes the cloud inference result in the queue for the local inference result for the same data item.