classification approach
Recently Published Documents

The recent development and miniaturization of hyperspectral sensors embedded in drones has allowed the acquisition of hyperspectral images with high spectral and spatial resolution. The characteristics of both the embedded sensors and drones (viewing angle, flying altitude, resolution) create opportunities to consider the use of hyperspectral imagery to map and monitor macroalgae communities. In general, the overflight of the areas to be mapped is conconmittently associated accompanied with measurements carried out in the field to acquire the spectra of previously identified objects. An alternative to these simultaneous acquisitions is to use a hyperspectral library made up of pure spectra of the different species in place, that would spare field acquisition of spectra during each flight. However, the use of such a technique requires developed appropriate procedure for testing the level of species classification that can be achieved, as well as the reproducibility of the classification over time. This study presents a novel classification approach based on the use of reflectance spectra of macroalgae acquired in controlled conditions. This overall approach developed is based on both the use of the spectral angle mapper (SAM) algorithm applied on first derivative hyperspectral data. The efficiency of this approach has been tested on a hyperspectral library composed of 16 macroalgae species, and its temporal reproducibility has been tested on a monthly survey of the spectral response of different macro-algae species. In addition, the classification results obtained with this new approach were also compared to the results obtained through the use of the most recent and robust procedure published. The classification obtained shows that the developed approach allows to perfectly discriminate the different phyla, whatever the period. At the species level, the classification approach is less effective when the individuals studied belong to phylogenetically close species (i.e., Fucus spiralis and Fucus serratus).

Drones have the potential to revolutionize malaria vector control initiatives through rapid and accurate mapping of potential malarial mosquito larval habitats to help direct field Larval Source Management (LSM) efforts. However, there are no clear recommendations on how these habitats can be extracted from drone imagery in an operational context. This paper compares the results of two mapping approaches: supervised image classification using machine learning and Technology-Assisted Digitising (TAD) mapping that employs a new region growing tool suitable for non-experts. These approaches were applied concurrently to drone imagery acquired at seven sites in Zanzibar, United Republic of Tanzania. Whilst the two approaches were similar in processing time, the TAD approach significantly outperformed the supervised classification approach at all sites (t = 5.1, p < 0.01). Overall accuracy scores (mean overall accuracy 62%) suggest that a supervised classification approach is unsuitable for mapping potential malarial mosquito larval habitats in Zanzibar, whereas the TAD approach offers a simple and accurate (mean overall accuracy 96%) means of mapping these complex features. We recommend that this approach be used alongside targeted ground-based surveying (i.e., in areas inappropriate for drone surveying) for generating precise and accurate spatial intelligence to support operational LSM programmes.