Maize leaf area index continuous monitoring based on time-series infrared images

Abstract: In situ measurements of leaf area index (LAI) are needed for monitoring crop growth conditions at the site level. In this study, we investigated the use of an infrared digital camera for canopy structural information extraction. Firstly, the camera (JAI industry camera, model AD-080CL, resolution: 1024×768, sensor size: 1/3 CCD, FPS: 30, focal length: 12 mm) was placed 6 m above the canopy, which was equipped with an infrared/color imaging switcher. Digital images were taken every 30 mins from above the canopy looking downward vertically. A Gaussian distribution based threshold technique was used to separate green vegetation tissues from background soil and residue materials in order to derive the canopy vertical gap fraction from the digital photos. The distribution of intensity for the background pixels was narrow compared to that of green plant pixels, The rising side of the background elements distribution on the histogram could be well-fitted with a Gaussian distribution function. However, the falling side overlapped the rising side of the green leaf distribution, which made it deviate from a Gaussian distribution at the junction of the two population distributions, the intensity of shaded leaves was generally smaller than that of sunlit leaves, which may cause confusion between shaded leaves and background elements. In this process, one of the challenges was how to overcome various natural light conditions which sometimes strongly affect the profile of the crop images taken from outdoor. In order to eliminate ambiguity which sometimes strongly affects the segmentation of the crop images taken outdoor, we adopted infrared images instead of color images, reflectance of healthy green plants in infrared images (760 nm) was higher than soil background due to red-edge phenomenon, which enhanced the intensity of shaded leaves. The intensity valueμin histogram represents the value at which the frequency of the background elements was maximal (the first peak from the left side of the histogram), and TL represented the intensity value below which the proportion of pixels was not more than 2.5% of the total number of pixels. The threshold TR was the mirror of TL with respect toμ, assuming a Gaussian distribution for background pixels. The use of 2.5% to determine the width of distribution of background pixels was to exclude pixels with exceptional low intensity while keeping the error below an acceptable level. In canopy structure parameter analysis phase, assuming the foliage was azimuthally uniform and spatially randomly distributed, the relationship between canopy gap fraction and LAI was given by the Poisson distribution. A spherical leaf inclination distribution function (LIDF) was considered a good first-order approximation for crop canopies, in which case foliage projection coefficient for the plane perpendicular to solar zenith angle was equal to 0.5 at any direction. LAI was estimated from the vertical gap fraction obtained from digital photography looking vertically downward. The experiment was conducted in 2016, two varieties in different decades (60s and 90s), density was 45 000 plants/hm2, normal water and fertilizer management. The LAI of two varieties varied greatly during the whole growth period, especially after grain filling, LAI of 60s variety decreased rapidly, however modern variety had an excellent "stay-green" characteristic, which maintained LAI at a certain high extent. A performance comparison of AccuPAR and our method was carried out, 18 samples were collected with two methods, the results showed that the LAI obtained using digital photography was linearly related with the AccuPAR method, the coefficient of determination between two methods hit 0.94. Our method provides a whole growth period automatic LAI monitoring solution.