Smartwatches provide a useful feature whereby users can be directly aware of incoming notifications by vibration. However, such prompt awareness causes high distractions to users. To remedy the distraction problem, we propose an intelligent notification management for smartwatch users. e goal of our management system is not only to reduce the annoying notifications but also to provide the important notifications that users will swiftly react to. To analyze how to respond to the notifications daily, we have collected 20,353 in-the-wild notifications. Subsequently, we trained the convolutional neural network models to classify important notifications according to the users' contexts. Finally, the proposed management allows important notifications to be forwarded to a smartwatch. As experiment results show, the proposed method can reduce the number of unwanted notifications on smartwatches by up to 81%.

No abstract

We are studying in-orbit real-time object detection for remote sensing satellites. Due to the small object size of remote sensing images, it is hard to achieve high detection accuracy, especially for resource-constrained spacecraft computers. Lightweight object detection models such as YOLO and SSD are feasible choices to achieve acceptable detection speed on board. This study proposes an accuracy-improvement method for the lightweight neural networks with an upscaling ratio estimator without retraining the model. The estimator exploits a scaling ratio that determines how much the image should be resized. With our scaling estimator, we have achieved 10.09% higher accuracy than the original YOLOv4-Tiny models with a 40% detection speed overhead.

Object detection from remote sensing images has been performed on the ground. Recently, on-board object detection has been studied only to show its feasibility with single-stage detectors. However, highly accurate models such as two stage detectors are compute intensive so that they are too slow to run on power-constrained on-board computers. In this paper, we propose a speed-up method for two-stage detectors. Two-stage detectors extract features and ROIs(Region of Interest) in the first stage and then classify them at the second stage. This structure gives high accuracy but induces large inference latency. In remote sensing images from satellites, object size is small relative to the whole image. Based on this characteristic, we propose to exclude features related to the large objects in the first stage. To verify our concept, we have selected various R-CNN models as two-stage object detectors. We have implemented our methods on two NVIDIA Jetson boards. We have achieved 1.8x speed up in inference latency with 5% accuracy drop with the small object dataset.

No abstract