This paper describes a low-cost and flexible vegetation classification system using a small and lightweight unmanned aerial vehicle (UAV). Compared with traditional remote sensing systems such as airplanes and satellites, small UAVs can obtain very high (a few cm) resolution aerial images. They are unaffected by cloud cover because they can fly at a very low altitude. In addition, owing to their small size and light weight, UAVs are highly portable, have a low operational cost, and can be operated safely. However vegetation classification based on traditional pixel-based classification approaches cannot work well when aerial images with cm-level resolution are used. Furthermore, the area of the aerial image captured by a UAV is limited to a small region due to its low-altitude flight. We propose a new method for determining vegetation type from aerial images obtained by a small UAV based on superpixel segmentation and machine learning techniques. We developed a method for creating a wide-area high-resolution mosaic image from multiple aerial images obtained by the UAV and onboard sensor data such as GPS and inertial sensors. Superpixel segmentation based on mean shift was then used to divide the mosaic image into small regions. Finally, the vegetation type was classified using the support vector machine (SVM) training and classifying process. We conducted our experiments at Yawata moor in Hiroshima Prefecture to classify moor vegetation. We evaluated the success rate of our proposed method based on the results of these experiments, and concluded that the small UAV system is both effective and useful for vegetation classification.
Although swamps and wetlands are considered valuable ecosystems, it is very difficult to perform field surveillance of such ecosystems because of flooding. Therefore, to develop an automatic classification method for detecting the flowers of the endangered plant species in wetlands, we observed wetland plants using a multirotor-type unmanned aerial vehicle (UAV). We obtained digital images of 76 scenes from approximately 15m above the ground with a spatial resolution of 5mm. Since the flowers were bright red and the surrounding leaves were green, the flower pixels were extracted based on color and brightness. To eliminate misclassification, additional filters such as shape, size and surroundings were applied, and the flower pixels were successfully extracted from the images of the target areas. Owing to the accuracy of the classification method, the flowers were extracted at 90.6% recall compared with visual inspection. Thus, the effectiveness of the approach was confirmed.
Semi-arid North-central Namibia is susceptible to poor crop harvests due to flood and drought. To mitigate the risk on the food security caused by flood and drought, rice cultivation could be introduced into the seasonal ponds (oondombe), which is one of topographical features in this region. The successful introduction of rice cultivation will depend on a good understanding of the hydrology of water storage in oondombe; however, few data exist. We therefore undertook spatiotemporal monitoring of oondombe water storages by integrating satellite remote sensing with measurements using structure-from-motion multi-view stereo (SfM-MVS) with an unmanned aerial vehicle (UAV). SfM-MVS is a recently developed technique that enables precise, simple, and inexpensive measurement of topography. First we conducted UAV surveys at 16 oondombe to generate a regression relationship between oondombe water extent and oondombe water storage volume. Then we observed daily oondombe water extent using several different sources of long-term satellite data (AMSR-E, AMSR2, MODIS, Landsat ETM+) interpreted with the assistance of recent data-fusion technique (database unmixing). Finally, we applied the regression relationship to the satellite data measurements of water extent and obtained estimates of oondombe water storage for the period 2002 to 2015 at three test sites in Namibia. The estimated oondombe water storage closely reflected seasonal change and year-to-year variation in flood and drought status. The accuracy of UAV measurements was several centimeters in the vertical direction and 10 cm in the horizontal direction. Comparison among oondombe water storage estimates with three different spatial resolution revealed that measurement with insufficient spatial resolution may lead to overestimation of water storage. This study not only revealed valuable data about oondombe water storage in this region, but also proposed a new approach for spatiotemporal hydrological monitoring over wide areas that merits additional research.
The remote sensing of ocean color by satellite is a useful technology for identifying water constituents which is highly repeatable, rapid, and covers a wide range. However, difficulties arise when we apply it to coastal waters due, for instance, to their optically complex constituents and low spatial resolution. In particular, the backward scattering coefficient, which governs the ocean color together with the absorption coefficient, is extremely difficult to measure. Recently, we developed a bypass method to determine the backward scattering coefficient with an accuracy of 3%. This method can help us to develop an ocean color algorithm for coastal waters. We also developed a multi-band imaging radiometer, which is compact and light weight and can be mounted on a multicopter. Our radiometer has 12 channels at wavelengths between 400nm and 800nm, and the signal-to-noise ratios of all channels can be adjusted to almost the same level with the help of neutral density filters. We made several test flights to check the performance of the radiometer in Suruga Bay. Although we only performed a simple analysis, the results showed that: 1) we could clearly detect high-density particle areas; 2) the influence of the atmosphere on the radiance measurements is negligible in the case of low-latitude observation; and 3) the observed spectral radiance exhibited the typical spectral shapes of Case2 Water.
Recently, food security has become a major world-wide concern. The development of a stable food supply system is necessary. Therefore, the practice of precision agriculture has received a great deal of attention. The management of crop growth is one of the most important aspects of precision agriculture, and it requires the accurate monitoring of the spatial distribution of the leaf area index (LAI) of target crops at the field scale. In recent years, unmanned aerial vehicles (UAVs) have been used to monitor the spatial distribution of the LAI of target crops at the field scale. The use of UAVs for LAI monitoring is expected to contribute to the efficiency of farming. However, no method for accurately estimating the spatial distribution of LAI using UAV has been established yet. In this study, several empirical regression models for estimating the LAI of paddy rice from several vegetation indices derived from UAV images were compared to help establish a precise estimation method of the spatial distribution of the paddy rice LAI at the field scale. As a result, although the estimation accuracy was not very high, this study indicated that vegetation indices that can consider the effect of vegetation density and whose estimation accuracy of LAI does not depend on the data used for calculating regression equations, such as the time-series change index of plant structure (TIPS), had higher potential to estimate the spatial distribution of the LAI of paddy rice than other vegetation indices that cannot consider the effect of vegetation density.
Surface ruptures associated with the 2014 Kamishiro fault earthquake (Mj 6.7), appeared along the Itoigawa-Shizuoka Tectonic line active fault system in the northern part of Nagano Prefecture, central Japan. We photographed it with digital cameras mounted on an unmanned aerial vehicle (UAV). Digital surface models (DSMs) were generated from the acquired photographs by applying SfM-MVS technology. The UAVs used in this study were the F450 and Phantom 2 manufactured by DJI Inc., and the cameras were the GR model manufactured by RICOH Inc. and were attached to each UAV for aerial photography. The ground control points required for generating DSMs using SfM-MVS analysis were measured using an RTK-GNSS (Leica GPS900), and the topographic profiles used for the accuracy assessment of the DSMs were measured in situ using a total station (Leica TCR705) and digital auto level (SOKKIA SDL50). As a result, we were able to create DSMs and ortho-photographs at the resolution of a few centimeters. The accuracy was assessed by comparing the topographic profiles measured by the total station and leveling with those generated by the DSMs. Validation against the nine topographic profiles revealed that the DSM had a relative height error of 4.0cm with an average standard deviation. Taking photographs from a UAV is one of the quickest and most cost-effective methods to record detailed surface topography. Generating a DSM of surface ruptures using UAV photography with SfM-MVS is particularly advantageous because ruptures will change their features quickly, and surface topographic variations of 10 cm or less cannot be recognized in field observations.
Possibilities of rice paddy monitoring using UAV were examined by SfM and filtering techniques in this study. A paddy farm in Nibutani, Biratori, Hokkaido was observed using DJI Inspire-1 drone with filtered on-board camera on 28 June 2015. DSM derived from UAV-SfM was indicated to be the useful to estimate rice plant height. Imagery analysis of Fire filtered drone on-board camera was discovered to be a promising new approach for rice paddy growth monitoring.
Regulations of UAV is not caught up with the progress of the evolving technology. Problem is happening a lot, because there is no uniform rules. Safety of the UAV and increasing interest on privacy protection, it is progressing rulemaking safe operation of the UAV at inside and outside the country. In this paper, we will introduce the trends of the current legislation in November 2015.
Red tide of eutrophic lake (Togo-ike) in Tottori prefecture were observed by an UAV (Unmanned Aerial Vehicle) observation system in order to establish a monitoring method for simply and speedy evaluation of water quality of the lakes. From the relationship between color information contained in reflected light from the lake and the water qualities of the lake such as Chlorophyll-a, the distribution map of the water quality was able to be displayed. The method would be able to be applied to the easily water quality evaluation of the lake.
Optimum timing for tea harvesting has been determined by farmers from visual observations based on their experience and knowledge. The automation of this determination process using imagery monitoring supports tea farmers on their cost and time in the field works. In this study, we attempted to develop a simple method to extract near-infrared signal from using filtered on-board camera with a drone (DJI Inspire1) in order to utilize for monitoring of the tea status. As a result, we recognized that the filtered image derived from Fire color filter (Rosco #19) was effective to extract near-infrared information over pre-harvesting second crop tea garden. Furthermore, we proposed new normalized vegetation index (FFDVI: Fire Filtered Drone Vegetation Index) using this extracted NIR band.
This paper presents an algorithm to select the effective bands for producing NDVI (Normalized Difference Vegetation Index) image using Hyperion hyperspectral data. The training data set of vegetation areas is prepared by extracting the digital number values from NIR (Near infrared) and RED bands. Exploratory factor analysis for the training data set indicates that the first and the second factor loadings correspond to NIR and RED bands, respectively, which can also be divided into two groups by those factor loadings. Based on the results, “Band 98 and Band 30” can be selected corresponding to maximum values of the first and the second factor loadings, respectively. As another approach, “Band 95 and Band 29” are selected with “maximum values of communalities”. Furthermore, “Band51 and Band30” are also selected corresponding to the central wavelengths of RED and NIR wavelength range in OLI(Operational Land Imager), respectively. Based on the aerial photograph, false color image and vegetation map, the comparative experiments for selected bands-based NDVI maps indicate that “communality” is superior to “factor loading” in selecting effective bands for NDVI mapping to avoid over- and under-estimating the values of NDVI, as well as to assure the robustness against the noise in Hyperion hyperspectral data.