Crop Density Estimation using Unmanned Aerial Vehicle Images

INTRODUCTION

we use input resources such as seeds, water, nutrients or pesticides to improve the efficiency or production of crops. The basics purpose of all these that we reduce heterogeneity spacing among plants in every row or maximum the density estimation among plants. For this we identify or introduce a model to find number of plants in each row. Also, it is used to identify the strength of any plant or crop.

Objectives

My main objectives to achieve in this research are:

• Due to a large variety of crops, it is very difficult for us to find out the specific number of plants, so, our main purpose is to develop a method like “Density estimation” with the help of modern technology like computer vision and UAV (Unmanned Aerial Vehicle)
• The main purpose of this study is to introduce such a system which is based on imagery of high resolution that can measure the population density of wheat plants
• Third objective is measuring crop density is very boring, labor intensive and costly to reduce it

In drone we use mobile to capture images then we further imaging processing using computer vision tool box in MATLAB. Comprehensive set of workflow-apps and algorithms of reference-standard can be determined with the help of image processing toolbox. Work flow applications are for analysis, algorithm development, visualization and image-processing. With help of this, we can perform 3D image processing, geometric transformations, image enhancement, and image registration and noise reduction. First, we have an image which original or aerial image. Then we apply image processing using computer vision toolbox of MATLAB. After apply filter we can identify feature detection or extraction of any plant or crop further classification of extraction images and finally distribute plant spacing or number of plants per row.

Any field that we use in our project is a square area that we use to capture images of any crop than we use it for further processing. Like we use this field to identify the crop density. Crop density means to identify the number of plants per row or area. Generally, density is a ratio of mass or volume.

The current methods used are based on visual plant counting in the fields over a predefined sampling area. They are tedious, time consuming and prone to human errors. Further, the soil conditions, particularly in case of rainfall or frost, may limit the time slots when walking in the field is possible without damaging the crop. Therefore, it is difficult to measure the plant density using such manual method when applied to large fields for documenting the spatial variability, or when applied to large phenotyping experiments with several hundred to several thousands of microplots to be characterized. It is thus necessary to develop alternative methods to provide timely and accurate quantification of the plant density. Precision agriculture (PA) is also defined as site of specific crop management (SSCM). By rational use of the balance resources or inputs it permits permission to the farmer too costly like Chemicals used only where or when they are essential. In further it is vital or necessary component of the system to promotion of the environmental stewardship.

Precision agriculture (PA) is not skill, it is a controlling idea to respond to three-dimensional inconsistency but PA technology makes it possible. The aim of Precision agriculture is to manage, examine and classify three dimensional and historical inconsistency with all facts of yield production in a field for best productivity, caring, sustainability and profitability for protection of environment and land resources because precision agriculture is a multi-disciplinary method. Once we obtain the variations than we see the reason behind these variation. Then try to remove or solve this variation by using the agricultural inputs or location bases within agriculture field. For obtaining maximum production it’s necessary to protect crop from diseases.

During recent decades several sensor techniques have been tested for crop density estimation. They can be divided in two categories based on the measurement principle, used; i.e. contact sensors and non-contact sensors. Contact sensors measure the deviation of a movable component of the sensor by virtue of its contact with the crop. The amplitude or frequency of this deviation is then registered and used as a measure for the crop density.

Unmanned Aerial Vehicles (UAV) images are typically collected via a flight mission with low altitude and large overlapping rate, whose spatial resolution is generally less than 10 cm. In contrast to the meter or decimeter-scale satellite image or aerial photographs, such UAV images are quite suitable for applying to detection of the occurrence of newly grown leaves over a tree crown or forest canopy.

As mentioned previously, newly grown leaves that randomly spread over tree crowns within a forest stand are quite small in size and less populous in the number of pixels particularly at the early and late stages in annual growing season. Success in detecting new leaves using remotely sensed data can help to determine the length of growing season and therefore potentially be of benefit in identifying or measuring how trees adapt or react to environmental change. So, the objectives of this paper were to investigate:

• The possibility of detecting new foliage from spectra-limited UAV bitmap images using hyperspectral target detection techniques,
• The performance of active and passive target detection techniques in differentiating the new foliage and their background targets, and
• The possibility of successful detection of new foliage using decimeter scale remote sensing images

UAV (unmanned aerial vehicle) is used for the purpose of to reach the target area that we use to wand see or also use it to destroy the target area of enemy because unmanned aerial vehicle contains sensors, violet weapons, spreader design and target designators. For manned aircraft system design the support life system and safety requirements for pilot protection in aircraft, while in unmanned aerial vehicle systems no pilot needs just controlling it on standing the earth and these systems are more accurate or efficient because where human reach is not possible, we use these vehicles.

However, none of these investigations specifically addressed the estimation of plant density. Advances in digital photography providing very high-resolution images, combined with the development of computer vision systems, offer new opportunities to develop a non-destructive high through put method for plant density estimation.

The objective of this study is to develop a system based on high resolution imagery that measures wheat plant population density at early stages. The methods used to acquire the RGB images and the experimental conditions are first presented. Then the pipeline developed to process the images is described. Finally, the method is evaluated with emphasis on its accuracy and on its corresponding domain of validity.

The normalized difference vegetation index (NDVI) can be used as sampling graphical indicator that could be used in order to analyze the distant sensing measurement, usually, but not essentially, formed a space platform, and evaluate whether the targeted object being perceived comprises lively green vegetation or maybe not.

The normalized difference vegetation index (NDVI), which was basically derived from remote satellite (sensing) data, was carefully connected to drought conditions. In order to determine the density of green vegetation on the square of a land, the discrete colors of various wavelengths of near-infrared and visible sunlight reproduced by the plants are being observed [2].

The values of NDVI changed between + 1.0 and − 1.0. The healthy vegetation had high near-infrared and lowered red-light reflectance that could produce higher values of NDVI. The mounting quantity of the positive values of NDVI specifies the increase within quantity of green vegetation. The values of NDVI near negative and about zero values indicated non-vegetated structures like barren surfaces (soil and rock), clouds, ice, snow and water. In this paper, the NDVI data was again rescaled to 0–255 in order to store the results within unspecified 8-bit data [2].

In order to investigate the density of green color on the specific patch of land, investigators should observe the discrete colors of near-infrared and visible sunlight (wavelengths) reflected through the plants. It could be seen with the help of prism; several altered wavelengths made the spectra of sunlight. When this light strikes on objects, definite wavelength of sunlight spectrum was absorbed, while all other wavelengths were reflected. The main pigment (chlorophyll) within leaves of plant powerfully absorbed visible light (ranging from 0.4-0.7 µm) for being used in the process of photosynthesis. The main cell structure of the plant leaves, on the other side, powerfully reflect the near-infrared light (from 0.7-1.1 µm). If a plant has a greater number of leaves, then, more quantity of these wavelengths is being affected.

Plant density and its non-uniformity drive the competition among plants as well as with weeds. They need thus to be estimated with small uncertainties accuracy. An optimal sampling method is proposed to estimate the plant density in wheat crops from plant counting and reach a given precision.

Crop density is a key agronomical trait used to manage wheat crops and estimate yield. Visual counting of plants in the field is currently the most common method used. However, it is tedious and time consuming. The main objective of this work is to develop a machine vision-based method to automate the density survey of wheat at early stages. RGB images taken with a high resolution RGB camera are classified to identify the green pixels corresponding to the plants. Crop rows are extracted and the connected components (objects) are identified. A neural network is then trained to estimate the number of plants in the objects using the object features. The method was evaluated over three experiments showing contrasted conditions with sowing densities ranging from 100 to 600 seeds. Results demonstrate that the density is accurately estimated with an average relative error of 12%. The pipeline developed here provides an efficient and accurate estimate of wheat plant density at early stages.

A field is defined in agriculture an area of the land used for agricultural purpose, nurturing of crop or also used as cattle. A field may also define or used unsown or unplanted [3].

Many farms are used for growing vegetables, which are used for the purpose to provide food for human as well as animals and also provide the shelter for wildlife. Many farms have border and used for the variety of plants or animals and this causes to reduce the crop yield.

Technology is rapidly changing the methods with which wildlife is being monitored. Unmanned aerial vehicles (UAVs) are one such example of new technology that allows biologists to take to the air to monitor wildlife, allowing for more cost-effective wildlife monitoring. UAVs allow for very high-resolution data acquisition in both the spatial and temporal domain. UAVs have been used in several civilian disciplines for research and monitoring: agriculture, forestry, and biodiversity monitoring including wildlife. The use of UAVs in wildlife studies is relatively recent and has focused more on the possibility of species detection.

The saturation problem associated with the use of NDVI for biomass estimation in high canopy density vegetation is a well-known phenomenon. Recent field spectroscopy experiments have shown that narrow band vegetation indices computed from the red edge and the NIR shoulder can improve the estimation of biomass in such situations.

In some causes we see that at the same area crop density is different it is because of many factors are include that causes crop density variations like: climate, land situation, amount of water, amount of fertilizer and many others factors in favors or against the crop density. As we know that our country economy based on agriculture therefore it is an important to increase the crop growth and health. Crop density estimation helps us to determine the crop health very clearly.

Today Precision agriculture is widely used. By using this term, you have ability to take decision how to reduce waste or protect the environment and maximize the profit. We use PA (Precision agriculture) by spraying we use input resources such as seeds, water, nutrients or pesticides to improve the efficiency or production of crops. The basics purpose of all these that we reduce heterogeneity spacing among plants in every row or maximum the density estimation among plants. For this we identify or introduce a model to find number of plants in each row. Also, it is used to identify the strength of any plant or crop [4].

In some causes we see that at the same area crop density is different it is because of many factors are include that causes crop density variations like: climate, land situation, amount of water, amount of fertilizer and many others factors in favors or against the crop density. As we know that our country economy based on agriculture therefore it is an important to increase the crop growth and health. Crop density estimation helps us to determine the crop health very clearly. Crop density is considering an important factor that supports in determining the crop yield. As we know that crops are extent along a very huge area. Manually or field methods for measuring crop density is very boring, labor intensive and costly. For large amount of crop or non-uniformity of plants it tough to find out number of plants so, the main objective of this work is to develop a method “density estimation” using modern technology such as UAV (Unmanned Aerial Vehicle) and computer vision.

Review of Literature

Mahesh et al. [5] investigated a method for getting the maximum yield of combine harvester, constantly intake of biomass was very important within threshing system was needed to be maintained. In order to achieve this, combined driver must be maintained by adjusting its speed continuously in accordance with the amount of biomass used on the front side of header. This presents the high pressure on the operator, who had used different other processes like cleaning and threshing.

Thorp et al. [6] describe that estimation of stand density of corn plant was best by the use of imagery collected at early reproductive growth stages of corn and the later vegetative stages. Among different types of cases which was plot, spatial resolution and image date, the basic constituents of reflectance mostly correlated with plant’s stand density were become able in order to classified within four prominent types, denoted in the form of types (A, B, C, and D). Type’ A’ basic components compared all wavelengths of visible red along all available wavelengths of near infrared. Type B basic components distinguished green wavelengths (nm) plus short wave near-infrared (nm) along red wavelengths ranging from 675nm to 693 nm with longer wave near-infrared of about 852 nm. Type C basic components included green wavelengths ranging from 528nm to 546 nm and near-infrared wavelengths ranged from 717nm to 855 nm. Type D basic components contrasted green/blue wavelengths ranged from 498nm to 507 nm along red edge of about 717 nm. Remote sensing could be accurately used in order to estimate stand density of corn plant at mid time of season as long as existence of plant stand variability and variability due to other various factors was minimal.

Moritz et al. [7] further proposed that systems like automatic feed-rate control can be used to enlighten the basic purpose of operator with the adjustment of driving speed in accordance with the quantity of biomass entered in the straw elevator. Currently, the incoming quantity of biomass was determined by torque on the drive belt header or the feeder.

Saeys et al. [8] according to his crop density means rate of feed was usually determined when material of crop was cut and loaded within machines. That disadvantages the control system operating automatically as compared to human operator which was able to view the density of crop in front of machines and can be able to estimate volume. But speed controllers working automatically always operated too late, a little bit, and can proceeds towards sub-optimal performance. Having an aim of increased comfort and speed, it was expected that a little information on the biomass quantity some meters in front of header would help in order to enhance the automatic control system.

Parker [9] described that due to crop density estimation or irregularity a competition occurs among weeds as well as in all plants. Due to this phenomenon, we needed accuracy to minimize the uncertainties. In order to estimate the plant’s density by counting the plants and accurate precision, optimal sampling method is developed. Crop density was the basic agronomical trait, which is used to manage the wheat crops and its yield. Plant’s visual counting within field is the most common method, which is in use. But it was a time consuming and tedious method. The basic purpose of this work was to develop a visual machine-based method in order to automate the density survey of wheat at its early stage. RBG images can be obtained by using high resolution camera in order to identify the green pixels of the plants. Rows of crop can be extracted, and its connected components can be identified by this method. After this, a neutral network trained in order to estimate the total plants in the features of the object.

Mutanga et al. [10] proposed that in the past few papers, density of plants and their non-uniformity was used to derive the competition within different plants along with their weeds. There was a need of their estimation with little accuracy. Optimal sampling method was usually used to estimate the density of plants in wheat crop from counting towards precision. For producing maximum amount of capacity for combined harvesting, a constant need of biomass was important within threshing systems, which was needed to be maintained.

Bendig et al. [11] described the problem of saturation was faced using the NDVI for different biomasses estimated to be in high canopy density vegetation. It was a well know phenomenon. Different techniques of spectroscopy were used for narrower band vegetation indices, which were computed from red edge. The NIR shoulder can be improved for the estimation of amount of biomass in such kind of situations.

Shi et al. [12] described that the variability within-row spacing and location of corn plant had been verified to have an important correlation with biomass and grain yield. They had included various models for yield prediction which were used in order to guide applications of variable-rate fertilizer within mid-season. Prototype sensing system has been developed in order to measure spacing and corn plant location on-the-go based on LiDAR technology on ground. The system moved along rows of crop with a sensor of ground LiDAR scanning at the lower bottom section of every corn plant. The probability of identification of corn stalk was enhanced because every stalk appeared in different scans from various types of view angles of that sensor. The firstly version of prototyping system had been developed at early stages and resulted towards lower detecting accuracy. Within paper, a better version of the prototyping system was developed with evaluation results of considerable additional field. This system had been improved within term of the data processing algorithm and data acquisition platforms, specially, the stalk recognition and scanned registration procedures in order to reduce the errors in misidentification.

Liu et al. [13] describe that crop density was main agronomical trait, which was used in order to manage the wheat crops with estimated yield. Visual-counting of various plants within field was presently the commonly used method. But it was time consuming and tedious. The main purpose of the work was the development of the machine vision-based method in order to mechanize the density-survey of plants of wheat at their early stages. RGB images, which have taken by RGB camera of high resolution were classified in order to recognize the green pixels conforming to the plants. Crop Rows of crops were extracted and their connected components were identified. After that, neural network was trained in order to estimate the total numbers of plants in the components by using features of the object. The method was appraised over total three experiments viewing contrasted conditions having sowing densities between the range of 100 to 600 seeds per square meter.

Lin et al. [14] described that the recent decades, different techniques regarding sensors had used for estimation of crop density. They had been divided into two categories based upon measurement of used principle i.e. non-contact sensors and contact sensors. Contact sensor was used to measure the deviation of that component of sensor, which was movable by virtue of its link with the crops. The frequency and amplitude of deviation was registered and to be used for measurement of crop density.

Thapa et al. [6] proposed that the most recent and fast development for computer vision algorithms and sensors gives different opportunities needed to be estimated plant density by using non-intrusive methods. Mostly studies on estimation of plants density were using non-contact measurements of ground level, mainly focused on large size plants like maize that can be regularly spaced. Mostly techniques used were based on identification of plants by using LiDAR systems and RGB images.

Zheng et al. [15] proposed in this paper another crop was taken which was rice from whole population approximately half world feeding this crop. From this situation we were able to say that in food security rice production plays a vital role. In this situation it’s very necessary or important to protect or care crops. In this paper new or different improve methods of UAV was introduced or applied.

Abu Bakar et al. [16] proposed that population density of rice plant was a basic indicator in order to determine the rate of fertilizer application and crop setting. It was therefore important that the density of population was monitored in order to ensure that an appropriate crop management decision was taken. The conventional method in order to determine population of plant was by physically counting the whole number of rice plants tillers within frame of 25 cm x 25 cm square. Sampling was done by arbitrarily choosing various locations in a plot to achieve tiller counting. This sampling process was costly, labour intensive and time consuming. An alternate fast estimating process was developed in order to overcome this type of issue. This process relies on calculating the ambit or outer circumference of the confined rice plants in a frame of 25 cm x 25 cm square in order to determine the exact number of tillers within square frame.