Apparatus and Infield Evaluations of a Prototype Machine Vision System for Cotton Plant Internode Length Measurement

摘要

An on-the-go infield machine vision system was developed to measure internode length (i.e., the distance between main stem nodes) of cotton (Gossypium hirsutum L.) plants, which is a significant indicator of past water stress. An infield vehicle was developed to convey the machine vision system automatically across the crop canopy. This paper presents an evaluation of the devised system under a range of operational and field conditions. On average, the vision system's internode length detection rate ranged from 12 measurements per 100 plants for compact plants when the sun was directly overhead with respect to the camera, to 64 measurements per 100 plants for vigorous plants when the sunlight was perpendicular to the camera view angle. Imaging at night time using 850 nm near-infrared illumination resulted in internode length detection rates not significantly different from the most reliable daylight conditions. Both across-row and along-row operations of the system were evaluated, with the system yielding internode length measurements for groundspeeds up to 0.20 m/s for along-row operation. Visual occlusion of the main stem nodes by foliage and variations in natural lighting conditions were observed to be the principal reasons for internode lengths not being detected successfully for every plant. However, from the success rates observed, it is concluded that the system has the practical capability to map internode length across a field and hence, identify spatial trends in internode length from which trends in plant water stress can be inferred.