journal article
Download Only Collection
Home
Ahmed, Oumer S.; Shemrock, Adam; Chabot, Dominique; Dillon, Chris; Williams, Griffin; Wasson, Rachel; Franklin, Steven E.
doi: 10.1080/01431161.2017.1294781pmid: N/A
The use of multispectral cameras deployed on unmanned aerial vehicles (UAVs) in land cover and vegetation mapping applications continues to improve and receive increasing recognition and adoption by resource management and forest survey practitioners. Comparisons of different camera data and platform performance characteristics are an important contribution in understanding the role and operational capability of this technology. In this article, object-based classification accuracies for different cover types and vegetation species of interest in central Ontario were examined using data from three UAV-based multispectral cameras. Five land-cover classes (forest, shrub, herbaceous, bare soil, and built-up) were determined to be up to 95% correct overall with calibrated multispectral Parrot Sequoia digital camera data compared to independent field observations. The levels of classification accuracy decreased approximately 10–15% when spectrally less capable consumer-grade RGB sensors were used. Multispectral Parrot Sequoia classification accuracy was approximately 89% when more detailed vegetation classes, including individual deciduous tree species, shrub communities and agricultural crops, were analysed. Additional work is suggested in the use of such UAV multispectral and point cloud data in ash tree discrimination to support emerald ash borer infestation detection and management, and in analysis of functional and structural vegetation characteristics (e.g. leaf area index).
doi: 10.1080/01431161.2016.1225182pmid: N/A
An autonomous unmanned aerial vehicle (UAV)-based aerial remote-sensing system was developed for multispectral imaging of agricultural farms. This system consists of on-board and ground-station subsystems. The on-board subsystem was equipped with body and wings, eight DC brushless motors and eight speed controls, an inertial measurement unit (IMU), a global positioning system (GPS), a rechargeable three-cell lithium-polymer battery, a high-resolution multispectral camera, camera mount, and autopilot intelligent navigation system. The ground station was equipped with a radio control, TV monitor, laptop, and flight monitor software. In addition, a telemetry system was used to communicate between the on-board and ground-station subsystems. To investigate the performance of the UAV, several flight tests were carried out in wheat farms, and some technical features of the system were investigated. The acquired imageries were processed and evaluated. The spatial resolution of the imagery obtained from a height of 10–250 m was 3.6 –95 mm/pixel, respectively. Based on the results, the UAV remote-sensing system proved to be extremely promising for monitoring the temporal changes in the studied farm. The supervised classification map with 94% overall accuracy and kappa coefficient of 0.9 emphasized the conclusions.
Cummings, Anthony R.; Karale, Yogita; Cummings, Garvin R.; Hamer, Esan; Moses, Persaud; Norman, Zacharias; Captain, Victor
doi: 10.1080/01431161.2017.1295487pmid: N/A
Swidden agriculture, or the continuing agricultural system in which clearings are cropped for shorter periods than they are fallowed, landscapes have been described as ‘difficult-to-map’ because they host a high variety of land-cover types. Consequently, satellite-borne remotely sensed data have not proven overwhelmingly successful in detecting and measuring change within these landscapes. We utilize data derived from an optical sensor carried on-board an unmanned aerial vehicles (UAVs) to measure the change on a swidden agriculture plot over a 4 month period in Guyana. UAVs were built with indigenous farmers who were trained in their operation to collect data over the swidden plot. A two-class classification was developed to quantify the change in both cultivated and naturally occurring vegetation. We found that non-vegetation surfaces rapidly decreased over the 4 months, declining from 79.42% in June to 69.69% in September. Vegetation recolonization of the swidden crop was particularly the cassava crop planted by the Makushi farmer. While our analysis was completed over a single swidden plot, our work demonstrated that UAVs could play a role in mapping swidden landscapes and change the perception that they are difficult to map. Local people involvement was critical to mapping their landscapes.
Hill, David J.; Tarasoff, Catherine; Whitworth, Garrett E.; Baron, Jackson; Bradshaw, Jacob L.; Church, John S.
doi: 10.1080/01431161.2016.1264030pmid: N/A
This study investigates the utility of an off-the-shelf, consumer-grade unmanned aerial vehicle (UAV) for invasive species mapping in a lacustrine fringe environment. Specifically, this work sought to determine whether such a UAV would be capable of creating accurate maps of the extent of patches of an invasive plant, yellow flag iris (Iris pseudacorus L.), more efficiently than could be accomplished by a traditional field survey, which is often considered in the literature to provide the most accurate maps. The study was conducted at two lakes in the central interior of British Columbia. The UAV used in this study was a DJI Phantom 3 Professional that can acquire images using the built-in 12.4 MP digital camera. This UAV was selected because it is representative of the type of aerial platform that is easily accessible to invasive plant managers in terms of cost, ease of use, and lack of legal restrictions. Three methods of mapping the yellow flag iris were compared: (1) field survey, (2) manual interpretation of the raw UAV-acquired imagery and the orthoimage created from these data, and (3) pixel-based classification of the orthoimage created from the UAV imagery using a random forest classifier. The results revealed that, at both lakes considered, manual interpretation of the UAV-acquired imagery produced the most accurate maps of yellow flag iris infestation, with a false-positive and false-negative classification rate of less than 1%.
Li, Wang; Niu, Zheng; Chen, Hanyue; Li, Dong
doi: 10.1080/01431161.2016.1235300pmid: N/A
The canopy structure of crops is a fundamental attribute of agro-ecosystems, providing efficient indications of the growing status, water stress, early detection of plant diseases, and yield estimation. In this study, we investigated the potential of point clouds generated from airborne laser scanning (ALS) and stereo images from an unmanned aerial vehicle (UAV) to characterize the structural complexity of maize canopies. The simultaneous collection of point clouds and field measurements were conducted on three sampling dates. A group of metrics that are often used in forest studies was calculated to quantify the structural complexity of canopies, which was further used to estimate the leaf area index (LAI) of maize. Stepwise linear regression models were established based on the metrics and LAI using the data sampled at single and mixed sampling dates, respectively. Results showed that significantly high correlations were found between the LAI values of maize and complexity metrics with a Pearson correlation coefficient (r) greater than 0.60. The leave one-out cross-validation (LOOCV) of LAI estimation showed that the highest robustness (RMSE = 0.16, rRMSE = 5.63%) was obtained by the model that was established from the overall data set, which explained 75% of the variation in the field-measured LAI. To conclude, the metrics of canopy structural complexity can be powerful predictors in the estimation of maize LAI based on our data set, which provides some new ideas for the study of precision agriculture using remote sensing.
Liu, Haiying; Zhu, Hongchun; Wang, Ping
doi: 10.1080/01431161.2016.1253899pmid: N/A
In this study, a big research progress has made in the research concerning leaf nitrogen content (LNC) nutritional spectral diagnosis on winter wheat at several growth stages, in which typical wave bands were put forward and quantitative models were constructed and validated. First, the unmanned aerial vehicle (UAV)-based hyperspectral data and the corresponding LNC data on winter wheat at several growth stages were obtained through experimenting in 2015, and the measured hyperspectral data and the LNC data were also obtained from the field-measured experimentation in 2014. Second, the spectral indices were calculated using UAV-based hyperspectral data and measured hyperspectral data, and the statistical regression models for diagnosing the LNC of different growth stages were constructed and analysed. Then, the correlation between the LNC and the spectral band is analysed. A method for selecting the typical bands of hyperspectral data responding to the LNC is proposed using spectral correlation as the basis. The UAV-based hyperspectral bands sensitive to the LNC of winter wheat are determined using this method. Finally, the hyperspectral quantitative models for diagnosing the LNC at the four stages are established by multifactor statistical regression and Back Propagation (BP) neural network methods. By comparing the modelling and verifying the coefficient, the UAV-based quantitative hyperspectral models’ effectiveness and practicability are then validated. The modelling results show that the predicted values are very ideal in jointing stage, flagging leaf stage, and flowering stage, while it is slightly less in the filling stage. The BP neural network modelling results were generally better than the multiple linear regression modelling results. This demonstrates that the effectiveness and spectrum sampling precision of UAV-based hyperspectral data are believable.
Massarelli, Carmine; Matarrese, Raffaella; Uricchio, Vito Felice; Muolo, Maria Rita; Laterza, Maurizio; Ernesto, Leanna
doi: 10.1080/01431161.2016.1226528pmid: N/A
This study aims to identify asbestos-containing materials (ACM) through the use of innovative technology such as aerial hyperspectral sensors. The development of operational methodologies and ad hoc processing were also applied for the purpose of this study. The activity was part of the ICT Living Labs DroMEP project carried out by Water Research Institute of the National Research Council (IRSA-CNR) and Servizi di Informazione Territoriale S.r.l. (SIT Srl). This was funded by the Apulia Region to support the growth and development of specialized SMEs in offering digital content and services. Uncontrolled abandoned wastes pose a threat to the human health and ecosystem. The presence of harmful or dangerous substances released without any control can become a dangerous source of pollution. Many areas of the Apulia region generally, in southern Italy, are subjected to this type of phenomena because most often, these areas are not easily accessible to Authorities for the control and management of the territory. Land monitoring and characterization operations would be carried out in a very long time and would require significant financial resources and considerable effort if done by conventional methods. The project activities include the testing and integration of several technologies already available, but not engineered for specific purposes. The work has been focused on the development of a methodology with a defined and high reliability capable of identifying the presence of ACM in various piles of rubbish abandoned in agro-ecosystems. The developed methodology analyses the spectral behaviour of ACM highlighting and emphasizing certain features through the use of a procedure based on an if–then–else control structure. It also allows the selection of the most effective features to combine that significantly reduces the number of false positives.
Matese, Alessandro; Di Gennaro, Salvatore Filippo; Berton, Andrea
doi: 10.1080/01431161.2016.1226002pmid: N/A
Biomass is one of the most important parameters in order for the farmer to choose the best canopy management within the field and it can be estimated using plant canopy height. In combination with a non-vegetation ground model, plant height can be obtained by quantifying the height of a canopy using crop surface models (CSMs). A modified Mikrokopter Okto unmanned aerial vehicle (UAV) acquired high-resolution multispectral images (4 cm) and a processing chain was developed to construct a 3D digital surface model (DSM) for the creation of precise digital terrain models (DTMs) based on Structure from Motion (SfM) computer vision algorithms. The DTM was then subtracted from the DSM to obtain a canopy height model (CHM) of a vineyard. The results show a good separation of ground pixels from vine rows, but their elevations were not quite in accordance with the actual height of the vines due to a smoothing effect of the reconstructed CHM. A further comparison between CHM and a vigour map obtained from normalized difference vegetation index (NDVI) values showed a good correlation. A preliminary assessment of biomass volume was made using the average canopy height and vine row width for three different homogeneous classes. This is a preliminary study on how a 3D model developed by UAV images can be useful for a simple and prompt biomass evaluation.
Mesas-Carrascosa, F. J.; Clavero Rumbao, I.; Torres-Sánchez, J.; García-Ferrer, A.; Peña, J. M.; López Granados, F.
doi: 10.1080/01431161.2016.1249311pmid: N/A
Weed mapping at very early phenological stages of crop and weed plants for site-specific weed management can be achieved by using ultra-high spatial and high spectral resolution imagery provided by multispectral sensors on-board an unmanned aerial vehicle (UAV). These UAV images cannot cover the whole field, resulting in the need to take a sequence of multiple overlapped images. Therefore, the overlapped images must be oriented and ortho-rectified to create an accurate ortho-mosaicked image of the entire field for further classification. Because the spatial quality of ortho-mosaicked images mainly depend on the flight altitude and percentage of overlap, this paper describes the effect of flight parameters using a multirotor UAV and a multispectral camera on the mosaicking workflow. The objective is to define the best configuration for the mission planning to generate accurate ortho-images. A set of flights with a range of altitudes (30, 40, 50, 60, 70, 80, and 90 m) above ground level (AGL) and two end-lap and side-lap settings (60–30% and 70–40%) were studied. The spatial accuracy of ortho-mosaics was evaluated taking into consideration the ASPRS test. The results showed that the best flight setting to keep the spatial accuracy in the bundle adjustment was 70–40% overlap and altitudes AGL ranging from 60 to 90 m. At these flight altitudes, the spatial resolution was quite similar, making it possible to optimize the mission planning, flying at a higher altitude and increasing the area overflow without decreasing the ortho-mosaic spatial quality. This study has relevant implications for further use in detecting weed seedlings in crops.
Showing 1 to 10 of 65 Articles