Multispectral-based leaf detection system for spot sprayer application to control citrus psyllids

A leaf detection module was developed for integration with a spot-sprayer system for controlling citrus psyllids. Since psyllids feed only on young citrus leaves, the module was designed to detect young leaves for selective spraying. The module comprised a four-band (570, 670, 750, and 870 nm) active optical sensor, a control box, and a data logger. Commands for operating the sensor and classifying leaves were embedded in a Java program run from a computer. The classification algorithm involved calculating different vegetation indices (using the sensorʹs spectral reflectance data) and implementing either the Euclidean Distance (ED) or the Matching Measures (MM) classifier. The module was tested under dynamic conditions using young-leaf-detection efficiency (ηYLD) and leaf-discrimination efficiency (ηLD) as performance criteria. ED classifier performed better than MM. Both ηYLD and ηLD decreased, with increasing variability, as target distance (TD) increased from 55 to 85 cm. TD of 55 cm gave the best performance (ηYLD = 100%, ηLD = 96.1%, and repeatability error, %eRmax = 0%). The system test comparing target length at three travel speeds for a leaf target, 15-cm long across the sensorʹs scanning path, confirmed that detected target length increased with travel speed. This was attributed to the diverging angle of the light beamed on the target, the effective sensing area, and the processing time difference between readings when the target first enters the sensorʹs field of view and when it has left. Overall, these results show good potential for the leaf detection module to be coupled with a spot sprayer.