9:00 AM - 10:30 AM
[ACG37-P11] Detection of a swarm of migratory locusts by a space-borne hyperspectral sensor
Keywords:Locust swarm detection, Migratory locust, Hyperspectral sensor, PRISMA, Locust plague
Locusts, such as the migratory locust (Locusta migratoria) and the desert locust (Schistocerca gregaria), are one of the most common pests. They often form huge swarms and damage crops. The large-scale feeding damage has been reported in many areas in the world. In a recent example, a large outbreak of desert locusts in 2020 caused damage over wide areas from the eastern Africa to the southern Asia. In Japan, two large outbreaks of migratory locusts occurred at the Kansai International Airport during the first island construction in 1994 and the second one in 1997. The numbers of locusts were estimated to be 13.38 million individuals on the first island and 38.84 million on the second one.
Locusts are usually in a "solitarious phase," and they do not swarm and avoid each other except mating. However, as locusts’ population density increases, the number of contact among individuals increase, and their phase changes to a "gregarious phase.” Locusts in the gregarious phase form swarms and improve their ability to fly.
To address the locust plagues, the distribution of locusts is currently monitored by human eyes, and this makes it difficult to continuously monitor vast areas with high resolution. Therefore, this study proposes a new method to detect gregarious locusts using a space-borne HyperSpectral (HS) sensor, where an HS sensor denotes an optical sensor which can measure spectrum at hundreds consecutive wavelengths.
The spectrum of the ground surface where gregarious locusts are there must be the mixture of the ground surface spectrum without locusts and the locusts spectrum. We utilized the spectrum measured by the space-borne HS sensor, PRISMA, as the former one. We measured the latter one using PicaL and HySpex SWIR-640 in the laboratory experiments. The both of them are the spectrum in the wavelengths between 400 nm and 2500 nm. The spectrum of the ground surface with gregarious locusts are calculated in each locusts coverage ratio (LCR), the ratio of the locust area to the ground area, in advance and used as the Look-Up Table (LUT).
We then simulate HS images of the ground surfaces with gregarious locusts, pseudo-HS images, using the ground surface PRISMA spectrum without locusts and the locusts spectrum because we have no such image.
To retrieve LCR of the pseudo-HS images by use of the LUT, we applied four spectral similarity measures: Euclidian Distance (ED), Spectral Angle Matching (SAM), Spectral Information Divergence (SID), and the Cross-Correlogram Spectral Matching (CCSM). The results showed that the estimation of LCR using the SID was the most accurate.
The Area Under Curve (AUC) of the Receiver Operating Characteristic (ROC) curve using SID was calculated to evaluate our method for the gregarious locusts detection. The results showed that the AUC exceeds 0.9 when LCR is more than 3% (1%) in one pixel ( 100 pixels), 30 m × 30 m (900 m × 900 m).
Locusts are usually in a "solitarious phase," and they do not swarm and avoid each other except mating. However, as locusts’ population density increases, the number of contact among individuals increase, and their phase changes to a "gregarious phase.” Locusts in the gregarious phase form swarms and improve their ability to fly.
To address the locust plagues, the distribution of locusts is currently monitored by human eyes, and this makes it difficult to continuously monitor vast areas with high resolution. Therefore, this study proposes a new method to detect gregarious locusts using a space-borne HyperSpectral (HS) sensor, where an HS sensor denotes an optical sensor which can measure spectrum at hundreds consecutive wavelengths.
The spectrum of the ground surface where gregarious locusts are there must be the mixture of the ground surface spectrum without locusts and the locusts spectrum. We utilized the spectrum measured by the space-borne HS sensor, PRISMA, as the former one. We measured the latter one using PicaL and HySpex SWIR-640 in the laboratory experiments. The both of them are the spectrum in the wavelengths between 400 nm and 2500 nm. The spectrum of the ground surface with gregarious locusts are calculated in each locusts coverage ratio (LCR), the ratio of the locust area to the ground area, in advance and used as the Look-Up Table (LUT).
We then simulate HS images of the ground surfaces with gregarious locusts, pseudo-HS images, using the ground surface PRISMA spectrum without locusts and the locusts spectrum because we have no such image.
To retrieve LCR of the pseudo-HS images by use of the LUT, we applied four spectral similarity measures: Euclidian Distance (ED), Spectral Angle Matching (SAM), Spectral Information Divergence (SID), and the Cross-Correlogram Spectral Matching (CCSM). The results showed that the estimation of LCR using the SID was the most accurate.
The Area Under Curve (AUC) of the Receiver Operating Characteristic (ROC) curve using SID was calculated to evaluate our method for the gregarious locusts detection. The results showed that the AUC exceeds 0.9 when LCR is more than 3% (1%) in one pixel ( 100 pixels), 30 m × 30 m (900 m × 900 m).