Abstract
Change Detection (CD) is a hot remote sensing topic where the change zones are highlighted by analyzing bi-temporal or multi-temporal images. Recently, Deep learning (DL) paved the road to implement various reliable change detection approaches that overcome traditional CD methods limitation. However, high performance DL based approaches have explosion number of parameters that demanded extensive computation and memory usage in addition to large volumes of training data. To address this issue, we proposed a teacher-student setting for remote sensing imagery change detection. To distill the knowledge from the over-parameterized Siamese teacher network, we proposed tiny student network that was trained using the obtained categorical distribution of probability from the teacher paired Softmax output at high temperature. Practical Swarm Optimization (PSO) was applied in order to optimally configure student architecture. Finally, ample experiments were conducted on LEVIR-CD dataset. Also, we introduced EGSAR-CD dataset, which contains of a large set of bi-temporal SAR images with 460 image pairs ( 256 ×256). Experiment results indicate that we can reach up to 5.4× reduction rate in number of parameters with loss of accuracy between 5% and 6% on the LEVIR-CD and EGSAR-CD datasets utilizing self-knowledge distillation.
Highlights
Change Detection (CD) is a crucial technique to effectively monitor earth surface changes using bi-temporal or multi-temporal data for the same geographic area
It is difficult to provide a specific explanation for why the margins differ on the two datasets, we argued that this inconsistency is attributable to differing spatial resolutions, inter-class similarities, intra-class variation, and inherent dataset characteristics
This paper introduces for the first time the knowledge distillation framework in remote sensing CD to tackle the limited hardware capacity in deployment
Summary
Change Detection (CD) is a crucial technique to effectively monitor earth surface changes using bi-temporal or multi-temporal data for the same geographic area. Examples of CD applications include urban expansion monitoring, resources and environment monitoring, disaster assessment, agriculture, and forestry [1]. The rapid generation of refined sensors and platforms help remote sensing (RS) community to achieve efficient automatic CD methods. CD is defined as a systematic process that composed of five sequential steps: 1) acquire suitable data, 2) preprocessing 3) extract suitable features, 4) CD algorithm, 5) performance evaluation. CD methods are classified into four categories [2] namely: 1) image algebra, 2) classification, 3) feature-based, 4) machine learning-based. Image algebra methods employ image difference and ratio
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
