Scene classification based on the bag-of-visual-words and Doc2Vec models for high-spatial resolution remote-sensing imagery

Delineating urban functional use from points of interest data with neural network embedding: A case study in Greater London

“…By contrast, the Doc2Vec model, an extension of the Word2Vec model developed by Le & Mikolov (2014), uses the neural network to train additional vectors for paragraphs, which can be utilised to specifically avoid the aforementioned problem and in directly training vectors for urban areas directly. Although there have been some applications of the Doc2Vec model in the urban context, few of them focus on detecting urban functional use with POI data (Wang et al, 2017;Li et al, 2019).…”

Adaptive bag‐of‐visual word modelling using stacked‐autoencoder and particle swarm optimisation for the unsupervised categorisation of images

“…Although Global Image representation via the Bag of Visual Word (BOVW) has been popular over the last two decades [28,29,30,31,32,33,34,35], and has been recognised to be most appropriate for Unsupervised Image categorisation process [20,36,37], the need to quantise a large number of image features into Visual Words using the K-Means algorithm during the BOVW codebook development creates a heavy a number of computational problems [21,24,25,38,39], and often yields Visual Words that do not guarantee optimum classification performance. Therefore, towards reducing the number of image features to be handled during BOVW Codebook Development and to allow, this section reviews some previous works related to the application of Deep Feature Learning to Image Representation and Vector quantisation in BOVW Image modelling.…”