EVALUATION OF RGB VEGETATION INDICES DERIVED FROM UAV IMAGES FOR RICE CROP GROWTH MONITORING

: The unmanned aerial vehicles (UAVs) are widely used for agricultural monitoring due to reduce the cost and time of crop monitoring via the acquisition of images with high spatial-temporal resolution. The normalized difference vegetation index (NDVI) is the most widely studied and used for mapping crop growth. A relatively expensive multispectral sensor is required to produce an NDVI map. The visible vegetation indices (VIs) derived from UAV images showed potential capabilities for predicting crop growth. The purpose of this paper is to evaluate the RGB indices to monitor the growth of the rice crop. The images were obtained from the study area by DJI PM4 multispectral UAV. The multispectral images were used to calculate NDVI as a reference vegetation index and different RGB indices were implemented and compared with the reference index. The results showed that RGB indices can be used as the vegetation index in the case of unavailable multispectral images.


INTRODUCTION
As a member of the cereal family, rice is the most widely used food item by a significant part of the world population, especially in Asia.As the world population increases, the demand for rice rises.So, the precise agriculture (PA) techniques have to be utilised to increase the crop yield.However, it is a challenge due to the fact that rice production is normally affected by major problems such as pests and diseases and the lack of supervision in the field.
Precision agriculture is an innovative and integrated approach, which enables farmers to make the best decisions for farm management.PA aims to improve the production of crops by maximizing crop yields.Technologies often used in PA include geographic information systems (GIS), remote sensing (RS) and geographic positioning systems (GPS), smart phones, and machine learning (Rudd et al., 2017a).A prerequisite for precision agriculture is data collection, which relies on the intelligent selection of sensors and their deployment in the field (Kelly et al., 2012;Rudd et al., 2017b).
Remote sensing technology is suitable for crop condition assessment based on the relationship between optical properties and biological parameters (Mac Arthur and Robinson, 2015).The plant phenotype, based on spectral reflectance information, relies on the optical properties reflected from the plant after multiple interactions (absorption, reflection, transmission, and emission) with plant tissues.Remote sensing data is used in various forms and sensors including satellites, and aerial in agriculture.Satellite or aerial remote sensing are widely used but have limitations such as resolution, availability and flexibility, complex image processing, and high costs (Mulla, 2013).These limitations can be overcome by substituting unmanned aerial vehicle (UAV) images.
Todays, UAVs are utilised in various fields of agriculture, including spraying, monitoring and management of farms, irrigation and disease detection, etc. (Abdullah et al., 2019;Hama et al., 2020;Starý et al., 2020).One of the main advantages of UAVs is capturing high-resolution RGB and multispectral images at low altitudes it is also light weight, easy to carry, lower cost and high manoeuvre.Spectral characteristics of plants in different wavelengths are influenced by chlorophyll, nitrogen, water content of plants, and morphological factors (Liu et al., 2013).Chlorophyll plays an important role in the spectral behavior of plants.It is found that when there is enough chlorophyll in the plant, the near-infrared band is most reflective when the plant is under stress.The plant growth is less than optimal and its natural productivity is interrupted which may cause the production of chlorophyll to decrease or stop.If the amount of chlorophyll decreases, the spectral behavior of the plant changes and causes less absorption in the blue and red bands (Atkinson et al., 2012).Therefore, the red and blue bands are also reflected along with the green band and gives yellow or red color to the vegetation.Remote sensing is used to quantitatively and qualitatively evaluation of vegetation and agricultural studies by examining vegetation indexes (VIs).The VIs are combinations of surface reflectance at two or more wavelengths designed to highlight specific vegetation characteristics (Liu et al., 2013;Zhou et al., 2021) .The VIs are calculated using operators and mathematical transformations on spectral reflectance in different spectral ranges of visible, near-infrared, middle and thermal.Therefore, by measuring the reflection of light after interacting with a plant, it is possible to determine the characteristics of a product based on the reflection values for different wavelengths of light .
The most popular VI is the normalized difference vegetation index (NDVI) which is widely used in vegetation measurement and agriculture due to its remarkable ability to detect vegetation compared to other indices in the literature (Li et al., 2019;Sun et al., 2017).The popularity of NDVI in vegetation sensing is due to the availability of infrared and red channels, because the reflectance of green and young vegetation increases in the near infrared channel and decreases in the red channel.Thus, a multispectral imaging sensor with infrared channel is required to exploit the NDVI for agricultural applications.However, multispectral sensors are expensive and require more processing and calibration.Currently, high resolution RGB cameras are more affordable and several plant indices based on color spectrum were used in recent years (Rabatel et al., 2011).Several research works have been conducted to use RGB UAV images instead of multispectral images in various precision agriculture applications including pest identification, plant identification, and yield estimation.
The capability of RGB and multi-spectral UAV images was evaluated to estimate nitrogen accumulation in rice crop (Zheng and Cheng, 2018).The results have shown that the use of the UAV images is a reliable approach to monitor N accumulation in the crop.Rice yield was estimated using multispectral imaged collected by UAV in northern Italy, showed that the UAV is a potential platform for investigating crop growth (Stroppiana et al., 2015).Also, the early growth stages of rice crops were investigated using the normalized difference index (Science, 2019).In another research, UAV derived plant indices such as canopy height and canopy cover were used to improve rice yield prediction under different nitrogen levels (Wan et al., 2019).In addition, the yield of rice crop was increased via the use of RGB UAV images in the early stage of crop (Norasma et al., 2018).In this research work, RGB and multispectral UAV images were used for rice plant growth monitoring.The obtained results from RGB and multispectral images were compared, and finally, the performance of the NDVI and RGB indices for rice crop growth were evaluated.

STUDY AREA
The study area is located in Babul city, the north of Iran, and the largest cities in Mazandaran province.The UAV images were collected during the ripening period of rice crop in 24 October 2020.The study area is shown in Figure 1.

Dataset
In this study, DJI Phantom 4 Multispectral (P4M) was used for imaging (Figure 2).It has a gimbal-based imaging system and includes RGB camera images and a multi-spectral camera array with five blue, green, red, red-edge and infrared cameras comprising 1600x1300 pixel resolution.More detailed information about the imaging sensor is provided in Table 1.Real-time, centimetre-accurate positioning data on images captured by all six cameras within DJI built-in system is used to align the RGB and multispectral images.Table 1.Specifications of the DJI PM4 UAV

Methods
The procedure of the proposed method is presented in Figure 3.

Geometric correction
The multi-camera sensors, such as DJI PM4 UAV camera, provide separated images for each band.Since the images have different perspective point, the raw images cannot be used without co-registration procedure.
In this paper, a feature-based registration technique, Speed-up Robust Feature algorithm (SURF), has been used to co-register the images collected via different band cameras.The extracted features have been used to estimate geometric transformation, and then, moving image was registered to the reference image with 1-pixel level of accuracy (Cruz et al., 2019).This technique was implemented in MATLAB and red band image was considered as the reference image and the rest of the images were registered to the reference.Figure 4 shows the red, green and blue bands before and after registration procedure.

Vegetation indices calculation
After co-registering the band images, three RGB vegetation indices including the Modified Green Red Vegetation index (MGRVI) (Bendig et al., 2015), Red Green Blue Vegetation index (RGBVI) (Bendig et al., 2015), Green Leaf Index(GLI) (Louhaichi et al., 2001), Modified excess green (MExG) (Burgos-artizzu et al., 2011) as well as NDVI have calculated.These indices have been widely used for crop health monitoring and were indicated in Table 2.
Table 2. Vegetation indices used in the study (R: Red, G: Green and B: Blue bands)

RESULT AND DISCUSSION
The maps of NDVI, MExG, RGVBI and MGRVI indices, R 2 and RMSE are extracted from the processed image and shown in

CONCLUSION
The use of drones for monitoring and management in agricultural products is very promising.This is because the images obtained from the uav have a very high spatial resolution.And they are flexible and easy to work with.On the other hand, plant indices are a useful and efficient tool for evaluating crops.Since multi-spectral sensors have a high cost, in this study they were used to evaluate RGB indexes for rice growth management.The results showed that it is possible to use RGB index for management in the field and prevent damage in crop production to a minimum.

Figure 2 .
Figure 2. DJI PM4 UAV Flight planning of 70 meters altitude with 65% forward-and side-overlaps was designed and image collection was carried out over the rice field at noon to minimize the shadow effect.Focal length 5.74 mm Image size 1600*1300 pixels Size of the sensor 4.87 x 3.96 mm Spectral bands Blue, green, red, near red edge and infrared

Figure 3 .
Figure 3. Procedure of the proposed method.