A green-from-brown test eliminated non-plant NDVI measurements, with subsequent plant discrimination achieved by classifying the calculated bivariate NDVI values for detected green plant material according to a predetermined Region-of-Classification (RoC) LUT. Depending upon the number of plant species N to be discriminated, the RoC LUT contained N parallelogram co-ordinate values that defined unique non-overlapping regions in NDVI635 and NDVI685 space. The discriminator output was an
!
N 30 vector denoted by Z Classified, that contained all of the NDVI635 and NDVI685 data pairs allocated to one of the N possible predetermined regions-of-classification, where the vector length was determined by the maximum possible number of measure-
!
ments on a per scan basis. Z Classified was then processed for plant strikes, where a strike was deemed to have happened when measured NDVI values from an unwanted plant fell within a specified RoC. To reduce the false alarm rate of the algorithm and to improve the PDU’s robustness and reliability, the strike detection algorithm was based on a running strike aggregate counter, which averaged the NDVI measurements of all of the individual beam strikes for a particular RoC. If a specified detection threshold Taggregate was exceeded, the strike was deemed valid and a plant spray signal generated.
A green-from-brown test eliminated non-plant NDVI measurements, with subsequent plant discrimination achieved by classifying the calculated bivariate NDVI values for detected green plant material according to a predetermined Region-of-Classification (RoC) LUT. Depending upon the number of plant species N to be discriminated, the RoC LUT contained N parallelogram co-ordinate values that defined unique non-overlapping regions in NDVI635 and NDVI685 space. The discriminator output was an!N 30 vector denoted by Z Classified, that contained all of the NDVI635 and NDVI685 data pairs allocated to one of the N possible predetermined regions-of-classification, where the vector length was determined by the maximum possible number of measure-!ments on a per scan basis. Z Classified was then processed for plant strikes, where a strike was deemed to have happened when measured NDVI values from an unwanted plant fell within a specified RoC. To reduce the false alarm rate of the algorithm and to improve the PDU’s robustness and reliability, the strike detection algorithm was based on a running strike aggregate counter, which averaged the NDVI measurements of all of the individual beam strikes for a particular RoC. If a specified detection threshold Taggregate was exceeded, the strike was deemed valid and a plant spray signal generated.
การแปล กรุณารอสักครู่..
