|
Autonomous Tributary Mapping using Multi-UAV System based on Semantic SLAM and Supervisory Control Theory |
|
Jeonghyeon Pak, Jaehwi Seol, Hyoung Il Son(Chonnam National University, Korea) |
| Remote sensing of the natural environment affected by abnormal weather is needed to manage natural systems and ecosystems. We use a singular value decomposition (SVD)-based tributary segmentation system that exploits the water absorption property of LiDAR. Considering the unstructured and diverging nature of tributaries, a multi-UAV system with reduced control complexity must be developed. Therefore, we propose a semantic simultaneous localization and mapping (SLAM)-based multi-UAV system for tributary mapping. This system is modeled using the hybrid automata method and controlled using a supervisory controller. |
|
|
Performance Evaluation of the Stem Cutting Mechanism using a Twisted String Actuation System |
|
Seongmo Choi, Myun Joong Hwang(University of Seoul, Korea) |
| This paper presents an evaluation of the cutting mechanism performance of the basket-typed gripper. The cutting mechanism is actuated by a twisted string. The bite force was measured by using a force sensor on the different sectors of the plate when the proposed mechanism cut the real stems. The experimental results show that it chopped the stems by applying up to 40.69 N of maximum cutting force and could cut the actual stems by exerting the various bite forces. |
|
|
Measurement and Analysis of Aerial Spray Characteristics using PointNet++ According to Flight Conditions |
|
Changjo Kim, Jaehwi Seol, Eunji Ju, Hyoung Il Son(Chonnam National University, Korea) |
| In this study, we propose a new method to measure and analyze the spray characteristics of unmanned aerial vehicles (UAVs) used in agriculture. This method utilizes mobile LiDAR-based data collection, spray segmentation using PointNet++, and k-mean clustering and Voronoi diagram algorithms. Spray characteristics were measured and analyzed under various flight conditions (height, speed, flight path). As a result of the study, it was confirmed that as the height and speed increased, the risk of drift spreading spray to non-target crops increased. This study can be used as an important basic study to increase the precision of sprays in the agricultural sector and reduce environmental and labor. |
|
|
Feature Vector Manipulation in the Latent Space for Classification Criterion Adjustment in CNN Models without Retraining |
|
Inhoon Jang, Jungin Kim, Changyu Park, Jimin Lee(Hankyong National University, Korea) |
| This paper covers a method to train a ConvNet to position feature vectors in a desired location on a high-dimensional latent space. Specifically, it proposes a method for training a contrastive learning model using triplet loss augmented with cosine similarity terms to position feature vectors at specific locations and orientations. Experimental results are presented to demonstrate the feasibility of the proposed approach. |
|