A Weakly Supervised Convolutional Network for Change Segmentation and Classification

A Survey of Un-, Weakly-, and Semi-Supervised Learning Methods for Noisy, Missing and Partial Labels in Industrial Vision Applications

“…Ref. [31] proposes the weakly supervised change detection method W-CDNet that can be trained with image-level labels. It uses a siamese architecture [80] to compare features from two different images.…”

Land Use Change Detection Using Deep Siamese Neural Networks and Weakly Supervised Learning

“…This is especially helpful for large-scale remotely sensed datasets [13]. Under this scenario, a DNN-based CD approach provides better performance on remotely sensed high-resolution images [1,16,21]. DNN based CD methods can be categorized as unsupervised [16], fully supervised [21] and weakly supervised approaches [1].…”

“…Under this scenario, a DNN-based CD approach provides better performance on remotely sensed high-resolution images [1,16,21]. DNN based CD methods can be categorized as unsupervised [16], fully supervised [21] and weakly supervised approaches [1]. Liu et al [16] developed an unsupervised CD algorithm using the pre-trained U-net architecture.…”

“…Under this scenario, a supervised CD model can be trained based on high-level annotations that indicate whether two images depict land change or not. This is known as a weakly supervised scheme [1,14]. Andermatt et al [1], proposed a weakly supervised CD approach using a U-net based Siamese architecture.…”

“…This is known as a weakly supervised scheme [1,14]. Andermatt et al [1], proposed a weakly supervised CD approach using a U-net based Siamese architecture. Similarly, Khan et al [14] have used a pre-trained DNN in conjunction with a directed acyclic graph (DAG) to learn patterns of change from image-level labelled training data.…”