Remote Sensing

We propose an algorithm based on linear prediction that can perform both the lossless and near-lossless compression of RF signals. The proposed algorithm is coupled with two signal detection methods to determine the presence of relevant signals and apply varying levels of loss as needed. The first method uses spectrum sensing techniques, while the second one takes advantage of the error computed in each iteration of the Levinson-Durbin algorithm. These algorithms have been integrated as a new pre-processing stage into FAPEC, a data compressor first designed for space missions.

 As a common dataset for change detection, its image can be divided into three parts: the image before the change, the image after the change, and the label image showing the changed area. This dataset is characterized by significant seasonal differences between bi-temporal image pairs, which makes up for some of the deficiencies in existing datasets. The labels for this dataset include some irregular changes, such as the appearance and disappearance of cars; but do not include seasonal changes, such as changes in the ground surface caused by snowfall.

Accurate detection and segmentation of apple trees are crucial in high throughput phenotyping, further guiding apple trees yield or quality management. A LiDAR and a camera were attached to the UAV to acquire RGB information and coordinate information of a whole orchard. The information was integrated by simultaneous localization and mapping network to form a dataset of RGB-colored point clouds. The dataset can be used for methods related to apple detection and segmentation based on point clouds.

In this project, an LSTM-based Model Predictive Controller (LSTM-MPC), with 200 neurons of each layer, is designed to have a highly efficient control on the temperature. The resulted dataset is attached, for further considerations.