Analysis of natural images processing for the extraction of agricultural elements

This work presents several developed computer-vision-based methods for the estimation of percentages of weed, crop and soil present in an image showing a region of interest of the crop field. The visual detection of weed, crop and soil is an arduous task due to physical similarities between weeds and crop and to the natural and therefore complex environments (with non-controlled illumination) encountered. The image processing was divided in three different stages at which each different agricultural element is extracted: (1) segmentation of vegetation against non-vegetation (soil), (2) crop row elimination (crop) and (3) weed extraction (weed). For each stage, different and interchangeable methods are proposed, each one using a series of input parameters which value can be changed for further refining the processing. A genetic algorithm was then used to find the best value of parameters and method combination for different sets of images. The whole system was tested on several images from different years and fields, resulting in an average correlation coefficient with real data (bio-mass) of 84%, with up to 96% correlation using the best methods on winter cereal images and of up to 84% on maize images. Moreover, the method’s low computational complexity leads to the possibility, as future work, of adapting them to real-time processing.     

Neural network and crop residue index multiband models for estimating crop residue cover from Landsat TM and ETM+ images

Crop residues on the soil surface provide not only a barrier against water and wind erosion, but they also contribute to improving soil organic matter content, infiltration, evaporation, temperature, and soil structure, among others. In Argentina, soybean (Glycine max (L.) Merill) and corn (Zea mays L.) are the most important crops. The objective of this work was to develop and evaluate two different types of model for estimating soybean and corn residue cover: neural networks (NN) and crop residue index multiband (CRIM) index, from Landsat images. Data of crop residue were acquired throughout the summer growing season in the central plains of Córdoba (Argentina) and used for training and validating the models. The CRIM, a linear mixing model of composite soil and residue, and the NN design, included reflectance and digital numbers from a combination of different TM bands to estimate the fractional residue cover. The results show that both methodologies are appropriate for estimating the residue cover from Landsat data. The best developed NN model yielded R² = 0.95 when estimating soybean and corn residue cover fraction, whereas the best fit using CRIM yielded R² = 0.87; in addition, this index is dependent on the soil and residue lines considered.     

Estimating leaf nitrogen concentration in heterogeneous crop plants from hyperspectral reflectance

The estimation of leaf nitrogen concentration (LNC) in crop plants is an effective way to optimize nitrogen fertilizer management and to improve crop yield. The objectives of this study were to (1) analyse the spectral features, (2) explore the spectral indices, and (3) investigate a suitable modelling strategy for estimating the LNC of five species of crop plants (rice (Oryza sativa L.), corn (Zea mays L.), tea (Camellia sinensis), gingili (Sesamum indicum), and soybean (Glycine max)) with laboratory-based visible and near-infrared reflectance spectra (300–2500 nm). A total of 61 leaf samples were collected from five species of crop plant, and their LNC and reflectance spectra were measured in laboratories. The reflectance spectra of plants were reduced to 400–2400 and smoothed using the Savitzky–Golay (SG) smoothing method. The normalized band depth (NBD) values of all bands were calculated from SG-smoothed reflectance spectra, and a successive projections algorithm-based multiple linear regression (SPA-MLR) method was then employed to select the spectral features for five species. The SG-smoothed reflectance spectra were resampled using a spacing interval of 10 nm, and normalized difference spectral index (NDSI) and three-band spectral index (TBSI) were calculated for all wavelength combinations between 400 and 2400 nm. The NDSI and TBSI values were employed to calibrate univariate regression models for each crop species. The leave-one-out cross-validation procedure was used to validate the calibrated regression models. Study results showed that the spectral features for LNC estimation varied among different crop species. TBSI performed better than NDSI in estimating LNC in crop plants. The study results indicated that there was no common optimal TBSI and NDSI for different crop species. Therefore, we suggest that, when monitoring LNC in heterogeneous crop plants with hyperspectral reflectance, it might be appropriate to first classify the data set considering different crop species and then calibrate the model for each species. The method proposed in this study requires further testing with the canopy reflectance and hyperspectral images of heterogeneous crop plants.     

Evaluating multispectral remote sensing and spectral unmixing analysis for crop residue mapping

Tillage practices can affect the long term sustainability of agricultural soils as well as a variety of soil processes that impact the environment. Crop residue retention is considered a soil conservation practice given that it reduces soil losses from water and wind erosion and promotes sequestration of carbon in the soil. Spectral unmixing estimates the fractional abundances of surface targets at a sub-pixel level and this technique could be helpful in mapping and monitoring residue cover. This study evaluated the accuracy with which spectral unmixing estimated percent crop residue cover using multispectral Landsat and SPOT data. Spectral unmixing produced crop residue estimates with root mean square errors of 17.29% and 20.74%, where errors varied based on residue type. The model performed best when estimating corn and small grain residue. Errors were higher on soybean fields, due to the lower spectral contrast between soil and soybean residue. Endmember extraction is a critical step to successful unmixing. Small gains in accuracy were achieved when using the purest crop residue- and soil-specific endmembers as inputs to the spectral unmixing model. To assist with operational implementation of crop residue monitoring, a simple endmember extraction technique is described.     

Using Bayesian networks with rule extraction to infer the risk of weed infestation in a corn-crop

This paper describes the modeling of a weed infestation risk inference system that implements a collaborative inference scheme based on rules extracted from two Bayesian network classifiers. The first Bayesian classifier infers a categorical variable value for the weed–crop competitiveness using as input categorical variables for the total density of weeds and corresponding proportions of narrow and broad-leaved weeds. The inferred categorical variable values for the weed–crop competitiveness along with three other categorical variables extracted from estimated maps for the weed seed production and weed coverage are then used as input for a second Bayesian network classifier to infer categorical variables values for the risk of infestation. Weed biomass and yield loss data samples are used to learn the probability relationship among the nodes of the first and second Bayesian classifiers in a supervised fashion, respectively. For comparison purposes, two types of Bayesian network structures are considered, namely an expert-based Bayesian classifier and a naïve Bayes classifier. The inference system focused on the knowledge interpretation by translating a Bayesian classifier into a set of classification rules. The results obtained for the risk inference in a corn-crop field are presented and discussed.     

Mitigating the effects of soil and residue water contents on remotely sensed estimates of crop residue cover

Crop residues on the soil surface decrease soil erosion and increase soil organic carbon and the management of crop residues is an integral part of many conservation tillage systems. Current methods of measuring residue cover are inadequate for characterizing the spatial variability of residue cover over large fields. The objectives of this research were to determine the effects of water content on the remotely sensed estimates of crop residue cover and to propose a method to mitigate the effects of water content on remotely sensed estimates of crop residue cover. Reflectance spectra of crop residues and soils were measured in the lab over the 400-2400 nm wavelength region. Reflectance of scenes with various residue cover fractions and water contents was simulated using a linear mixture model. Additional spectra of scenes with mixtures of crop residues and soil were also acquired in corn, soybean, and wheat fields with different tillage treatments and different water content conditions. Crop residue cover was linearly related to the cellulose absorption index (CAI), which was defined as the relative intensity of an absorption feature near 2100 nm. Water in the crop residue significantly attenuated CAI and changed the slope of the residue cover vs. CAI relationship. Without an appropriate correction, crop residue covers were underestimated as scene water content increased. Spectral vegetation water indices were poorly related to changes in the water contents of crop residues and soils. A new reflectance ratio water index that used the two bands located on the shoulders of the cellulose absorption feature to estimate scene water conditions was proposed and tested with data from corn, soybean, and wheat fields. The ratio water index was used to describe the changes in the slope of crop residue cover vs. CAI and improve the predictions of crop residue cover. These results indicate that spatial and temporal adjustments in the spectral estimates of crop residue cover are possible. Current mutispectral imaging systems will not provide reliable estimates of crop residue cover when scene water content varies. Hyperspectral data are not required, because the three narrow bands that are used for both CAI and the scene moisture correction could be incorporated in advanced multispectral sensors. Thus, regional surveys of soil conservation practices that affect soil carbon dynamics may be feasible using either advanced multispectral or hyperspectral imaging systems.     

Uncertainty analysis and evaluation of a complex,multi-specific weed dynamics model with diverse and incomplete data sets

Weed dynamics models are needed to test prospective cropping systems but are rarely evaluated with independent data (“validated”). Here, we evaluated the FlorSys model which quantifies the effects of cropping systems and pedoclimate on multispecific weed dynamics with a daily time step. We adapted existing validation methodologies and uncertainty analyses to account for multi-specific, multi-annual and diverse outputs, focusing on missing input data, incomplete and imprecise weed time series. Field data ranged from entirely monitored cropping system trials to annual snapshots recorded on farm fields by the French Biovigilance-Flore network. FlorSys satisfactorily predicted weed seed bank, plant densities and crop yields, at daily and multi-annual scales, at well monitored sites. It overestimated plant biomass and underestimated total flora density. Missing processes (photoperiod dependency in flowering, crop:weed competition for nitrogen) and inadequately predicted scenarios (weed dynamics in untilled fields, floras with summer-emerging species) were identified. Guidelines for model use were proposed.     

Harris Hawks Optimizer with Graph Convolutional Network Based Weed Detection in Precision Agriculture

Precision agriculture includes the optimum and adequate use of resources depending on several variables that govern crop yield. Precision agriculture offers a novel solution utilizing a systematic technique for current agricultural problems like balancing production and environmental concerns. Weed control has become one of the significant problems in the agricultural sector. In traditional weed control, the entire field is treated uniformly by spraying the soil, a single herbicide dose, weed, and crops in the same way. For more precise farming, robots could accomplish targeted weed treatment if they could specifically find the location of the dispensable plant and identify the weed type. This may lessen by large margin utilization of agrochemicals on agricultural fields and favour sustainable agriculture. This study presents a Harris Hawks Optimizer with Graph Convolutional Network based Weed Detection (HHOGCN-WD) technique for Precision Agriculture. The HHOGCN-WD technique mainly focuses on identifying and classifying weeds for precision agriculture. For image pre-processing, the HHOGCN-WD model utilizes a bilateral normal filter (BNF) for noise removal. In addition, coupled convolutional neural network (CCNet) model is utilized to derive a set of feature vectors. To detect and classify weed, the GCN model is utilized with the HHO algorithm as a hyperparameter optimizer to improve the detection performance. The experimental results of the HHOGCN-WD technique are investigated under the benchmark dataset. The results indicate the promising performance of the presented HHOGCN-WD model over other recent approaches, with increased accuracy of 99.13%.     

LANDSAT MSS crop classification performance as a function of scene characteristics

In order to utilize remote sensing fully to inventory crop production, it is important to identify and quantify the effects of the factors that affect the accuracy of LANDSAT classifications. The objective of this study was to investigate the effect of scene characteristics involving crop, soil and weather variables on the accuracy of LANDSAT classifications of corn and soybeans. Segments of multitemporally registered LANDSAT MSS data from two key acquisition periods sampling the U.S. Corn Belt were classified using a Gaussian maximum likelihood classifier. Field size had a strong effect on classification accuracy with small fields tending to have low accuracies even when the effect of mixed pixels was eliminated. Other scene characteristics accounting for variability in classification accuracy included proportions of corn and soybeans, crop diversity index, proportion of all field crops, soil order, soil drainage class, percentage of slope, long-term average soybean yield, maximum yield, relative position of the segment in the Corn Belt, weather and crop development stage.     

Sensitivity study of Radarsat-2 polarimetric SAR to crop height and fractional vegetation cover of corn and wheat

Increasing studies have been conducted to investigate the potential of polarimetric synthetic aperture radar (SAR) in crop growth monitoring due to the capability of penetrating the clouds, haze, light rain, and vegetation canopy. This study investigated the sensitivity of 16 parameters derived from C-band Radarsat-2 polarimetric SAR data to crop height and fractional vegetation cover (FVC) of corn and wheat. The in-situ measured crop height and FVC were collected from 29 April to 30 September 2013, at the study site in southwest Ontario, Canada. A total of 10 Radarsat-2 polarimetric SAR images were acquired throughout the same growing season. It was observed that at the early growing stage, the corn height was strongly correlated with the SAR parameters including HV (R² = 0.88), HH-VV (R² = 0.84), and HV/VV (R² = 0.80), and the corn FVC was significantly correlated with HV (R² = 0.79) and HV/VV (R² = 0.92), but the correlation became weaker at the later growing stage. The sensitivity of the SAR parameters to wheat variables was very low and only HV and Yamaguchi helix scattering showed relatively good but negative correlations with wheat height (R² = 0.57 and R² = 0.39) at the middle growing stage. These findings indicated that Radarsat-2 polarimetric SAR (C-band) has a great potential in crop height and FVC estimation for broad-leaf crops, as well as identifying the changes in crop canopy structures and phenology.     

Estimation of plant species by classifying plants and leaves in combination

Information on which weed species are present within agricultural fields is a prerequisite when using robots for site‐specific weed management. This study proposes a method of improving robustness in shape‐based classifying of seedlings toward natural shape variations within each plant species. To do so, leaves are separated from plants and classified individually together with the classification of the whole plant. The classification is based on common, rotation‐invariant features. Based on previous classifications of leaves and plants, confidence in correct assignment is created for the plants and leaves, and this confidence is used to determine the species of the plant. By using this approach, the classification accuracy of eight plants species at early growth stages is increased from 93.9% to 96.3%.     

基于颜色特征的杂草图像分割技术研究

为了解决目前杂草识别中受光照影响大、环境适应性差等问题,提出了基于颜色特征的分割算法。此算法在统计分析杂草和土壤背景各颜色因子的基础上,得到适于杂草图像分割的颜色分量,实现了复杂场景、光照条件下杂草区和背景区的分割。实验结果表明:R-G,2G-R-B,Hmean,Smean,Hmean Smean颜色特征对于杂草区和背景区的分割能够取得很好的效果,可广泛应用于田间杂草识别、树种识别、人脸识别等受光照变化影响较大的领域。     

Robust joint stem detection and crop‐weed classification using image sequences for plant‐specific treatment in precision farming

Conventional farming still relies on large quantities of agrochemicals for weed management which have several negative side‐effects on the environment. Autonomous robots offer the potential to reduce the amount of chemicals applied, as robots can monitor and treat each plant in the field individually and thereby circumventing the uniform chemical treatment of the whole field. Such agricultural robots need the ability to identify individual crops and weeds in the field using sensor data and must additionally select effective treatment methods based on the type of weed. For example, certain types of weeds can only be effectively treated mechanically due to their resistance to herbicides, whereas other types can be treated trough selective spraying. In this article, we present a novel system that provides the necessary information for effective plant‐specific treatment. It estimates the stem location for weeds, which enables the robots to perform precise mechanical treatment, and at the same time provides the pixel‐accurate area covered by weeds for treatment through selective spraying. The major challenge in developing such a system is the large variability in the visual appearance that occurs in different fields. Thus, an effective classification system has to robustly handle substantial environmental changes including varying weed pressure, various weed types, different growth stages, changing visual appearance of the plants and the soil. Our approach uses an end‐to‐end trainable fully convolutional network that simultaneously estimates plant stem positions as well as the spatial extent of crop plants and weeds. It jointly learns how to detect the stems and the pixel‐wise semantic segmentation and incorporates spatial information by considering image sequences of local field strips. The jointly learned feature representation for both tasks furthermore exploits the crop arrangement information that is often present in crop fields. This information is considered even if it is only observable from the image sequences and not a single image. Such image sequences, as typically provided by robots navigating over the field along crop rows, enable our approach to robustly estimate the semantic segmentation and stem positions despite the large variations encountered in different fields. We implemented and thoroughly tested our approach on images from multiple farms in different countries. The experiments show that our system generalizes well to previously unseen fields under varying environmental conditions—a key capability to deploy such systems in the real world. Compared to state‐of‐the‐art approaches, our approach generalizes well to unseen fields and not only substantially improves the stem detection accuracy, that is, distinguishing crop and weed stems, but also improves the semantic segmentation performance.     

Effects of remote sensing pixel resolution on modeled energy flux variability of croplands in Iowa

With increased availability of satellite data products used in mapping surface energy balance and evapotranspiration (ET), routine ET monitoring at large scales is becoming more feasible. Daily satellite coverage is available, but an essential model input, surface temperature, is at 1 km or greater pixel resolution. At such coarse spatial resolutions, the capability to monitor the impact of land cover change and disturbances on ET or to evaluate ET from different crop covers is severely hampered. The effect of sensor resolution on model output for an agricultural region in central Iowa is examined using Landsat data collected during the Soil Moisture Atmosphere Coupling Experiment (SMACEX). This study was conducted in concert with the Soil Moisture Experiment 2002 (SMEX02). Two images collected during a rapid growth period in soybean and corn crops are used with a two-source (soil+vegetation) energy balance model, which explicitly evaluates soil and vegetation contributions to the radiative temperature and to the net turbulent exchange/surface energy balance. The pixel resolution of the remote sensing inputs are varied from 60 m to 120, 240, and 960 m. Model output at high resolution are first validated with tower and aircraft-based flux measurements to assure reliability of model computations. Histograms of the flux distributions and resulting statistics at the different pixel resolutions are compared and contrasted. Results indicate that when the input resolution is on the order of 1000 m, variation in fluxes, particularly ET, between corn and soybean fields is not feasible. However, results also suggest that thermal sharpening techniques for estimating surface temperature at higher resolutions (∼250 m) using the visible/near infrared waveband resolutions could provide enough spatial detail for discriminating ET from individual corn and soybean fields. Additional support for this nominal resolution requirement is deduced from a geostatistical analysis of the vegetation index and surface temperature images.     

Monitoring soybean growth using L-, C-, and X-band scatterometer data

A ground-based fully polarimetric scatterometer operating at multiple frequencies was used to continuously monitor soybean growth over the course of a growing season. Polarimetric backscatter data at L-, C-, and X-bands were acquired every 10 min. We analysed the relationships between L-, C-, and X-band signatures, and biophysical measurements over the entire soybean growth period. Temporal changes in backscattering coefficients for all bands followed the patterns observed in the soybean growth measurements (leaf area index (LAI) and vegetation water content (VWC)). The difference between the backscattering coefficients for horizontally transmitted horizontally received (HH) and vertically transmitted vertically received (VV) polarizations at the L-band was apparent after the R2 stage (DOY 224) due to the double-bounce scattering effect. Results indicated that L-, C-, and X-band radar backscatter data can be used to detect different soybean growth stages. The results of correlation analyses between the backscattering coefficient for specific bands/polarizations and soybean growth data showed that L-band HH-polarization had the highest correlation with the vegetation parameters LAI (r = 0.98) and VWC (r = 0.97). Prediction equations for estimation of soybean growth parameters from the L-HH were developed. The results indicated that L-HH could be used for estimating the vegetation biophysical parameters considered here with high accuracy. These results provide a basis for developing a method to retrieve crop biophysical properties and guidance on the optimum microwave frequency and polarization necessary to monitor crop conditions. The results are directly applicable to systems such as the proposed NASA Soil Moisture Active Passive (SMAP) satellite.     

A modified linear extrapolation approach towards red edge position detection and stress monitoring of wheat crop using hyperspectral data

The inflection point of spectral reflectance of crop in the red edge region (680–780 nm) is termed as the red edge position (REP), which is sensitive to crop biochemical and biophysical parameters. We propose a technique for automatic detection of four dynamic wavebands, i.e. two in the far-red and two in the near-infrared (NIR) region from hyperspectral data, for REP estimation using the linear extrapolation method. A field experiment was conducted at the SHIATS Farm, Allahabad, India, with four levels of nitrogen and irrigation treatments to assess the sensitivity of REP towards crop stress. A correlation analysis was carried out between REPs and different biophysical parameters, such as leaf area index (LAI) and chlorophyll content index (CCI), recorded in each plot at 50, 70, and 90 days after sowing of wheat crop under the field experiment. The inter-comparison among different REP extraction techniques revealed that the proposed technique, i.e. the modified linear extrapolation (MLE) method, has a better ability to distinguish different crop stress conditions. REPs extracted using the MLE technique showed high correlations with a wide range of LAI, CCI, and LAI × CCI, being comparable with results obtained using the traditional linear extrapolation and polynomial fitting techniques. The behaviour of the new techniques was found to be stable at both narrower and broader bandwidth, i.e. 2 and 10 nm. A new red-edge-based index, i.e. area under REP (A_REP), was used to detect the cumulative stress over wheat crop by utilizing the REP and its rate of change information at different crop growth stages. A high coefficient of determination (R² = 0.89) was found between A_REP and dry grain yield (Q ha^?1) up to 50 Q ha^?1 of wheat crop, whereas, beyond this range the relationship was found to be diminishing.     

Errors associated with estimating vegetation water content from radar for use in passive microwave brightness temperature algorithms

The Soil Moisture Active Passive Validation Experiment 2012 was conducted as a pre-launch validation campaign for the Soil Moisture Active Passive mission over 6 weeks in June and July 2012. During this campaign, the Passive Active L-Band System (PALS) was flown at a low altitude, providing radar and radiometer measurements that were contained within a single agricultural field. The campaign domain consisted of 55 agricultural fields, where soil moisture was measured coincident to the PALS flight times and measurements of vegetation volumetric water content (VWC) and leaf area index (LAI) were measured weekly. The low-altitude flights allowed for the comparison between measured VWC and LAI for 11 fields to radar parameters derived from the radar backscatter. Only the correlation between the HV backscatter and the soybean VWC was considered strong (|r| > 0.7). All other correlations between the radar parameters and the VWC (or LAI) were moderate (0.3 < |r| < 0.7) or weak (|r|< 0.3). The established relationships between radar parameters and VWC were used in a forward radiation transfer model to estimate H-pol brightness temperature. It was found that the RMSE between the brightness temperatures estimated using the measured VWC was lowest when using the relationship between VWC and LAI (3.9 K for soybeans, 6.8 K for spring wheat, and 9.3 K when all crop data are combined). Despite a lower correlation, the RMSE associated with using the radar vegetation index relationship with VWC was less than when HV was used (7.9 K) for soybeans, which would result in an error in soil moisture estimation of just over 4%. The RMSEs for all other VWC and radar parameter relationships were greater than 10 K.     

ANFIS-based prediction of the compressive strength of geopolymers with seeded fly ash and rice husk–bark ash

In the present work, compressive strength of geopolymers made from seeded fly ash and rice husk–bark ash has been predicted by adaptive network-based fuzzy inference systems (ANFIS). Different specimens, made from a mixture of fly ash and rice husk–bark ash in fine and coarse forms and a mixture of water glass and NaOH mixture as alkali activator, were subjected to compressive strength tests at 7 and 28 days of curing. The curing regimes were different: one set of the specimens were cured in water at room temperature until 7 and 28 days and the other sets were oven-cured for 36 h at the range of 40–90°C and then cured at room temperature until 7 and 28 days. A model based on ANFIS for predicting the compressive strength of the specimens has been presented. To build the model, training and testing using experimental results from 120 specimens were conducted. The used data as the inputs of ANFIS models are arranged in a format of six parameters that cover the percentage of fine fly ash in the ashes mixture, the percentage of coarse fly ash in the ashes mixture, the percentage of fine rice husk–bark ash in the ashes mixture, the percentage of coarse rice husk–bark ash in the ashes mixture, the temperature of curing, and the time of water curing. According to these input parameters in the ANFIS models, the compressive strength of each specimen was predicted. The training and testing results in ANFIS models showed a strong potential for predicting the compressive strength of the geopolymeric specimens.     

基于图像处理技术的杂草特征提取方法研究

利用计算机视觉技术将杂草从背景中识别出来进行定位喷洒农药已成为精细农业研究的热点。选取颜色空间OHTA中I＇2分量作为特征量;利用基于遗传算法的自动阈值选取方法对特征分量巧进行阈值分割初步分离杂草与小麦;通过颜色聚类和形态滤波获得准确的杂草区域。实验结果表明：直接在彩色空间进行分割,可提高彩色图像的分割效果,利用该方法获得的杂草平均正确识别率达到90．47％。     

An evaluation of radar as a crop classifier

Results of an investigation to specify the parameters of a space-borne imaging radar for use in crop identification are discussed. The study relied upon scaltering data acquired with a groundbased radar which were degraded to simulate the performance of a system similar to the proposed Space Shuttle Orbiter Imaging Radar. Data acquired from fields sown in corn, milo, soybeans, wheat and alfalfa were employed. The results of this study suggest that for best classification accuracy, a K-band (approximately 14 GHz), dual polarized system viewing fields at an off nadir angle in the 40° to 60° range should be employed. However it is emphasized that to attain classification accuracies exceeding 90%, multi-date acquisition is required. As best as can be determined, four target revisits at an interval of ten days is adequate for 90% accuracy.