Natural Resources Canada (NRCan) requires the development and evaluation of specific machine learning methods, i.e., transfer learning and domain adaptation techniques, applied to optical imaging. More specifically, the work must meet the following objectives:
Changes in Spatial/Spectral Resolution (Different Sensors)
Develop methods to transfer perceptual knowledge captured in a trained deep convolutional neural network (DCNN) to other images from satellites that use different sensors and therefore have different spatial and spectral characteristics. Specifically, this goal seeks to develop approaches to take models developed for Worldview data and transfer them to datasets with spatial resolutions of between 1-2.5m with potentially similar spectral characteristics.
Change in Domain (Different Geographic Areas)
Develop methods to transfer perceptual knowledge trained on high resolution data in specific regions of Canada to areas that exhibit different geographic conditions and will build upon and use the models developed in objective 1.
The requirements are aimed at developing the two methods (Learning by Transfer and Non-Supervised Domain Adaptation) and evaluating the performance of these methods, through a set of exhaustive tests on data provided by NRCan. The methodology should allow for simple and easy reuse by NRCan. To do so, the service provider will have to provide NRCan with the test plan, the results and the code developed to perform the tests. In addition, the code provided to reuse the methodology will have to be delivered with an open license (MIT).
Suppliers' right to submit a statement of capabilitiesSuppliers who consider themselves fully qualified and available to provide the goods, services or construction services described in the ACAN may submit a statement of capabilities in writing to the contact person identified in this notice on or before the closing date of this notice. The statement of capabilities must clearly demonstrate how the supplier meets the advertised requirements.
