Determination of nitrogen status in Mavuno pastures through image analysis using machine vision technology

• Grant number: 20/09425-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: October 01, 2020
• End date: September 30, 2021
• Field of knowledge: Agronomical Sciences - Animal Husbandry - Pastures and Forage Crops
• Principal Investigator: Lilian Elgalise Techio Pereira
• Grantee: Ilnara Rodrigues Silva
• Host Institution: Faculdade de Zootecnia e Engenharia de Alimentos (FZEA). Universidade de São Paulo (USP). Pirassununga , SP, Brazil

Abstract

The efficient management of fertilization for the Brazilian pastures is a critical to the profitability and sustainability of the livestock pasture-based systems. However, the traditional tools for monitoring N content in pastures, based on laboratorial analysis, are expensive and time consuming. Thus, new imaging-based automatic analysis, most of them based on machine vision techniques, have been studied as a low-cost, fast and practical alternative to the traditional tools for plant's nutritional diagnosis. The main aim of the present proposal is to develop a mobile app, called leaf color chart (LCC), to monitor to monitor N status at real-time from field-taken images using a smartphone camera. The LCC is a simple tool used as an indicator for leaf color which is a proxy for its nitrogen content, being a technology widely used in China, Pakistan, Tailand and other countries. To implement an LCC it is necessary to define a chart of colors, which represents classes of N status of the plants of interest. Since there is no LCC developed for pastures until now, we first need to correctly identify how many colors will be necessary/adequate to represent different N status of the plants. A field experiment was carried out from February 2019 to April 2020 using a Brachiaria hybrid, known as Mavuno grass. To generate gradients of N content in the plants, four nitrogen rates were defined and applied after each cutting: no-nitrogen (N0), 15 (N15), 30 (N30) and 45 (N45) kg of N ha-1 using urea, and were distributed in randomized complete blocks with four replications (plots of 20 m²). When plots have reached the defined cutting target (40 cm pre-cutting), all measurements and images were taken. It was evaluated: Forage mass and the morphological composition and the leaf area index. At the time of cutting, the youngest completely expanded leaf (diagnostic leaf) from 20 tillers randomly chosen into the plot, placed on an image collecting table with an opaque white background, and images were obtained. The digital images were stored as 24-bit color images and saved in RGB color space in JPEG format. The leaves will compose a sample in which will be determined the total N content (%Na). The critical N concentration (%Nc) will be calculated, and a regression analysis (on a monthly or season basis), considering the relationships between N fertilization rates applied in the field, the forage accumulation and the resulting %Na, will provide the models to estimate the required N rates to attain the critical level and the maximum forage accumulation. Each image will be classified according its classes of N status previously identified. A data set corresponding the N status of each image, after pre-processed, and its respective values of the red, green, blue channels as well as hue, saturation and brightness will be subjected to a cluster analysis, to detect how many groups of colors can be separated from the data set. From each one of the resulting groups the colors will be recreated. The mobile application will contain the codes for image acquisition, scanning for the determination of RGB and HSB and a similarity algorithm (Euclidean distance or others to be tested) will be used to classify the samples collected in the field in relation to the colors defined in the LCC. Based on the identified LCC pattern, the mobile application will be capable of to recommend the fertilization rate to maintain the critical N content and to attain the maximum herbage accumulation.