Development approach and evaluation of the Nutrient Expert software for nutrient management in cereal crops

Meeting the demand for more food in the next 20–30 years requires intensifying cereal cropping systems and increasing current yields to about 70–80% of the genetic yield potential. A dynamic and robust nutrient management approach such as site-specific nutrient management (SSNM) will be essential to increase yields and optimize profits while maintaining the productivity of these intensive cropping systems. SSNM has increased yield and profit in rice, maize, and wheat in major cropping systems in Asia; but, crop advisors have found it complex and difficult to implement in the field. Nutrient Expert (NE) was developed to provide crop advisors with a simpler and faster way to use SSNM. NE enables crop advisors to develop SSNM recommendations using existing site information. Nutrient Expert for Hybrid Maize (NEHM) increased yield and profit of farmers in Indonesia and the Philippines. In Indonesia, NEHM increased yield by 0.9 t ha⁻¹, which increased profit by US$ 270 ha⁻¹ over farmer’s fertilizer practice (FFP). Compared with FFP, NEHM recommendations reduced fertilizer P (−4 kg ha⁻¹), increased fertilizer K (+11 kg ha⁻¹), and did not significantly change fertilizer N. In the Philippines, NEHM increased yield by 1.6 t ha⁻¹ and profit by US$ 379 ha⁻¹ compared with FFP. Compared with FFP, NEHM gave higher rates of all three nutrients (+25 kg N ha⁻¹, +4 kg P ha⁻¹, and +11 kg K ha⁻¹), which substantially increased fertilizer costs (US$ 64 ha⁻¹) but still increased profit by about six times the additional investment in fertilizer. NE accounts for the important factors affecting site-specific recommendations, which makes it a suitable starting point for developing nutrient management tools to reach more users.     

A hybrid method combining SOM-based clustering and object-based analysis for identifying land in good agricultural condition

Remotely sensed imagery is currently used as an efficient tool for agricultural management and monitoring. In addition, the use of remotely sensed imagery in Europe has been extended towards determination of the areas potentially eligible for the farmer subsidies under the Common Agricultural Policy (CAP), through interactive or automatic land cover identification. For accurate quantification and fast identification of agricultural land cover areas from the imagery, a hybrid method, which combines automated clustering of self-organizing maps with object based image analysis, and called SOM + OBIA, is proposed. Performance analysis on three test zones (using multi-temporal Rapideye imagery) indicates that for the basic land cover categories (forest, water, vegetated areas, bare areas and sealed surfaces), unsupervised classification with the proposed SOM + OBIA method achieves an identification accuracy comparable to the accuracy of the traditional interactive object oriented analysis, with considerably less user interaction.     

A feasibility study of the classification of Alpaca (Lama pacos) wool samples from different ages,sex and color by means of visible and near infrared reflectance spectroscopy

The usefulness of classifying the Alpaca wool samples according to their color, sex and location is associated with their economic value in the market, hence adequate methods for rapid classification are needed to assess the of wool value. This study evaluated the potential of the visible and near infrared (vis–NIR) spectroscopy combined with multivariate statistical analysis to classify Alpaca (Lama Pacos) fiber samples according to age (1 and 2–3-year-old), sex (Male and Female) and color (Black, Brown, LF and White). Samples (n = 291) were scanned in reflectance mode in the wavelength range of 400–2500 nm using a monochromator instrument (FOSS NIRSystems6500, Inc., Silver Spring, MD, USA). Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were used to classify fiber samples. Cross-validation was used for validation of classification models developed. Results showed that PLS-DA correctly classified 100% of fiber samples into ages, intermediate classification rates were obtained for color, while lower classification rates were obtained for the discrimination of wool samples according to sex. The results from this study suggested that vis–NIR spectroscopy in combination with multivariate data analysis can be used as a rapid method to classify Alpaca fiber samples according to age, sex and color.     

Analysis of hyperspectral scattering images using locally linear embedding algorithm for apple mealiness classification

Hyperspectral scattering images between 600 nm and 1000 nm were acquired for 580 ‘Delicious’ apples for mealiness classification. A locally linear embedding (LLE) algorithm was developed to extract features directly from the hyperspectral scattering image data. Partial least squares discriminant analysis (PLSDA) and support vector machine (SVM) were applied to develop classification models based on the LLE, mean-LLE and mean spectra algorithms. The model based on the LLE algorithm achieved an overall classification accuracy of 80.4%, compared with 76.2% by the mean-LLE algorithm and 73.0% by the mean spectra method for two-class classification (i.e., mealy and nonmealy) coupled with PLSDA. For the SVM models, the LLE algorithm had an overall classification accuracy of 82.5%, compared with 79.4% by the mean-LLE algorithm and 78.3% by the mean spectra method. Hence, the LLE algorithm provided an effective means to extract hyperspectral scattering features for mealiness classification.     

Nondestructive measurement of internal quality of Nanfeng mandarin fruit by charge coupled device near infrared spectroscopy

The soluble solids content (SSC) and total acidity (TA) are the major characteristics for assessing quality and maturity of Nanfeng mandarin fruits. The feasibility of charge coupled device near infrared spectroscopy (CCD-NIRS) combining with effective wavelengths selection algorithm used to measure SSC and TA nondestructively was investigated. The effective wavelengths to SSC and TA were chosen by interval partial least squares (iPLS) in the wavelength range of 600–980 nm. The predictive ability of SSC model used PLS regression was improved with r of 0.92 and RMSEP of 0.65 °Brix using effective wavelengths of 681.36–740.51 nm, 798.60–836.19 nm and 945.52–962.75 nm. The TA model was simplified with r of 0.64 and RMSEP of 0.09% using effective wavelengths of 817.57–836.19 nm, 909.85–927.60 nm and 945.52–962.75 nm. The experimental results demonstrated that the CCD-NIRS technique combining with iPLS algorithm was a feasible method to measure SSC and TA of Nanfeng mandarin fruits nondestructively.     

Evaluation of three-dimensional accelerometers to monitor and classify behavior patterns in cattle

Cattle behavior is potentially a valuable indicator of health and well-being; however, natural movement patterns can be influenced by the presence of a human observer. A remote system could augment the ability of researchers, and eventually cattle producers, to monitor changes in cattle behavior. Constant video surveillance allows non-invasive behavior monitoring, but logging the movement patterns on individual animals over long periods of time is often cost prohibitive and labor intensive. Accelerometers record three-dimensional movement and could potentially be used to remotely monitor cattle behavior. These devices collect data based on pre-defined recording intervals, called epochs. Our objectives were to (1) determine if accelerometers can accurately document cattle behavior and (2) identify differences in classification accuracy among accelerometer epoch settings. Video-recorded observations and accelerometer data were collected from 15 crossbred beef calves and used to generate classification trees that predict behavior based on accelerometer data. Postural orientations were classified as lying or standing, while dynamic activities were classified as walking or a transition between activities. Video analysis was treated as the gold standard and logistic regression models were used to determine classification accuracy related to each activity and epoch setting. Classification of lying and standing activities by accelerometer illustrated excellent agreement with video (99.2% and 98.0% respectively); while walking classification accuracy was significantly (P < 0.01) lower (67.8%). Classification agreement was higher in the 3 s (98.1%) and 5 s (97.7%) epochs compared to the 10 s (85.4%) epoch. Overall, we found the accelerometers provided an accurate, remote measure of cattle behavior over the trial period, but that classification accuracy was affected by the specific behavior monitored and the reporting interval (epoch).     

Agricultural applications of a low-cost infrared thermometer

Plant canopy temperature is used in many studies of plant/environment interactions and non-contact measurement is often made with radiometric surface thermometers commonly referred to as infrared thermometers. Industrial-quality infrared thermocouples are widely available and often used in agricultural research. While research on canopy temperature has provided management tools for production agriculture, the high cost of the industrial-quality infrared thermocouples has limited their adoption and use in production agriculture settings. Our objective was to evaluate a low-cost consumer-quality infrared thermocouple as a component of a wireless thermal monitoring system designed for use in a production agriculture setting. The performances of industrial-quality and low-cost consumer-quality sensors were compared under controlled constant temperature and under field conditions using both grass and cotton canopies. Results demonstrate that under controlled constant-temperature the two types of infrared thermocouples were “significantly the same” at 10 °C, 20 °C and 30 °C and “significantly not the same” at 40 °C and 50 °C. Across the temperature range tested, the consumer-quality infrared thermocouples temperature reading was closer to the thermocouple reading than the industrial-quality infrared thermocouples. A field comparison of industrial-quality and consumer-quality infrared thermocouple sensors monitoring a grass canopy and a cotton canopy indicated that the two types of sensors were similar over a 13–35 °C range. The measurement of temperature made with two types of sensors would not differ significantly. Based on these results we conclude that the lower-cost consumer-quality infrared thermometers are suitable for use in production agricultural applications.     

Comparison of segmentation algorithms for cow contour extraction from natural barn background in side view images

Computer vision techniques are a means to extract individual animal information such as weight, activity and calving time in intensive farming. Automatic detection requires adequate image pre-processing such as segmentation to precisely distinguish the animal from its background. For some analyses such as gait analysis, a side view perspective is recommended. When using a side view angle however, the background is more difficult to control – moving objects, such as other animals may negatively impact successful image segmentation. The objective of this research was to evaluate five different background segmentation algorithms on side view images when taken against a static background (a solid transportable wall) and a dynamic background (open air, without a wall).The experiments were conducted on a commercial robotic-milking dairy farm in Israel with a herd size of 70 Israeli Holstein cows. A side view image of cow’s gait was recorded after milking when the cows exited the milking area and returned to the cowshed. From the recording database, a random selection was made of 35 frames containing a static background (solid wall) and 20 frames containing a dynamic background (natural barn environment with other cows).Five segmentation algorithms were chosen and adapted from literature to extract the cow shape from the image. The output of three algorithms gave the cow’s full body shape two identified only the contour of the cow’s body. The algorithms were compared on their ability to correctly identify the cow’s back contour line. The performance of each algorithm was quantified by comparing its outputs to a golden standard of manually labelled cow pixels in the image.The introduction of a physical wall behind the cows (static background) significantly improved the foreground segmentation results (Mean Absolute Error (MAE) = 6.7 ± 5.7 pixels vs. 19.7 ± 9.1 pixels). The fourth algorithm, based on an edge detection on the background difference frame, gave the best cow back contour line segmentation results (b₀ = −0.4 ± 15.5 and b₁ = 1.00 ± 0.07). The fifth algorithm which is based on consecutive frame differences was less accurate than the other four methods which are based on the background frame differences (MAE = 16.0 ± 5.9 pixels vs. 4.1 ± 2.2 pixels, 4.3 ± 2.2 pixels, 5.6 ± 2.8 pixels and 3.7 ± 1.4 pixels respectively for the other four algorithms). The results show that the applied algorithms were not robust enough to work on side view images with dynamic backgrounds.     

Identification of grapevine varieties using leaf spectroscopy and partial least squares

Grapevine variety identification is a matter of great interest in viticulture, which is currently addressed by visual ampelometry or wet chemistry genetic analysis. This paper reports the development of a simple and automatic method of classification of grapevine varieties from leaf spectroscopy. The method consists of a classifier based on partial least squares that discriminates among grapevine varieties using a hyperspectral image of a leaf measured in reflectance mode. Hyperspectral imaging was conducted with a camera with 1040 wavelength bands operating between 380 nm and 1028 nm. The classifier was created using 300 leaves, 100 of each of the varieties Vitis vinifera L., Tempranillo, Grenache and Cabernet Sauvignon. Monte-Carlo cross-validation confirmed the classifier’s performance for the three varieties, which exceeded 92% in all cases. The proposed method has proven to satisfactory classify among grape varieties, but certainly a wider range of grapevine cultivars should be tested before it gets implemented for local sensing with the aim of providing the wine industry with a fast, automatic, environmentally friendly and accurate tool for grapevine variety identification.     

Automated system developed to control pH and concentration of nutrient solution evaluated in hydroponic lettuce production

Lettuce is one of the most widely consumed leaf vegetables. In hydroponic the growth depends upon the composition of nutrient solution. Due to its nutrient absorption, the conductivity and pH suffer continuous variations. This paper describes the development of a system completely managed by a lab-made software. It monitors the conductivity and pH throughout 24 h during the whole cycle of production. Also, allows adjust automatically any variation, through solenoid valves which dispense solutions of acid/base or nutrient. The efficiency of the proposed instrumentation was evaluated by simultaneously cultivation of same kind of lettuce (Vanda) in two different ways, hydroponics in greenhouse controlled with the developed devices, and grown conventionally in soil, adopted as referential. Agronomic and chemical parameters of commercial interest were analyzed for both crop, attesting the precocity in harvest (64 against 71 days) with reduced labor, better control and higher productivity, especially in fresh and dry matter of aerial parts, presenting 267.56 and 13.33 g plant⁻¹ respectively, using the developed system. The data sequence regarding the concentration of nutrients for the automated hydroponic system was similar to those obtained by the mentioned researchers, as follows: K > N > Ca > P > Mg > S > Fe > Zn > Mn > Cu. This similarity highlights the efficiency of controlling the parameters of conductivity and pH in the instrumental system applied to hydroponics, offering the producer an effective and viable alternative in the production of lettuce.     

Cough localization for the detection of respiratory diseases in pig houses

In this paper, an algorithm is proposed for localizing cough sounds by using the difference in arrival time of a sound signal between different microphones. To assess the accuracy of the proposed method, estimated positions of a reference sound were compared with real positions in various microphone configurations. All the configurations showed a good position estimation, with mean error between 1.5 and 0 m, and a maximum standard deviation on the error of 0.4 m. The algorithm was also applied on the localization of coughing pigs in a stable in field condition. Using this method it was possible to localize cough attacks of pigs. During a 3 h recording trial, sound was recorded using seven microphones in which 179 coughs were collected, originating from 19 cough attacks. After mapping the locations in the stable, three hazard zones could be identified. This information can be used for visualizing the spread of respiratory diseases and eventually contribute to the reduction of the use of antibiotics by means of selective and early treatment of single pens in stead of the whole compartment.     

Bruise detection on ‘Golden Delicious’ apples by vis/NIR spectroscopy

Visible and near-infrared reflectance spectra were used to distinguish the bruises from the intact surfaces on ‘Golden Delicious’ apples. Reflectance spectra of apples were acquired for both bruised and intact surfaces within the range of 400–1700 nm. The effective wavebands for detecting bruises were determined by analysis of the correlelogram. The wavebands around 545 and 1200 nm clearly show the time evolution of the bruised tissue. The mean-normalized reflectance difference between wavebands centred at 745 and 905 nm was also an effective discriminator for detecting old bruises. Afterwards, a quadratic discrimination analysis was performed based on the selected discriminators. The total classification error for the 1-day-old bruises was about 16.3%. A more extensive study was carried out to determine the effects of storage time on classification performance. The detection error was decreased with the elapse of storage time after bruising. However, it was difficult to remove the time influence on the classification accuracy if only depending on the intensity value of reflectance.     

Modeling of sows diurnal activity pattern and detection of parturition using acceleration measurements

This article suggests and assesses two different monitoring methods for detecting sows parturition using series of three-dimensions acceleration measurements previously classified into activity types. Two groups of sows are monitored: a first group (n = 9) provided with straw (S), and a second group (n = 10) where no straw is provided (NS); two types of activity are taken into account: high active behaviour (corresponding to feeding, rooting and nest building behaviours) and total active behaviour (including any active activity type). The first method suggests modeling sows’ diurnal pattern of activity using a saw-tooth function for the probability of being active and monitoring the series using a Dynamic Generalized Linear Model (DGLM). The second method is based on a cumulative sum of hourly differences of activity, from day-to-day. Both methods use a threshold value, optimized for each group, to detect the onset of farrowing. Best results in terms of sensitivity and specificity are observed for the cumulative sum method, using individual variance and monitoring high active (sensitivity = 100%; specificity = 100%) and total active behaviours (sensitivity = 100%; specificity = 95%). Results of the DGLM method indicate a sensitivity of 100% and a specificity of 89% in average for both group S and NS. Observing the occurrence of alarm times, the DGLM method allows (i) earlier detection of farrowing: 15 h before the onset of farrowing, for both groups, as compared to 9–12 for the other methods; and (ii) a better distribution of alarms, i.e. minimize the number of alarms occurring within the last 6 h before farrowing.     

A high quality low-cost digital microscope minirhizotron system

An easily reproducible system is presented for minirhizotron image acquisition based on a digital microscope and entirely built with low-cost components. The system weighs 2.28 kg, is connected to a portable computer through USB, and allows collection and storage of high-quality digital images without other components or power sources. Settings at 25× magnification provide pixel sizes of 25 and 12 μm respectively with a medium pixel density microscope (640 × 480 pixel), and a high pixel density microscope (1280 × 1024 pixel). This kind of system coupled with recent developments of robotics and automatic image analysis for minirhizotron frames allows to envisage fast, low-labour and low-cost root investigation methods with a high degree of automation.     

Potential of a terrestrial LiDAR-based system to characterise weed vegetation in maize crops

LiDAR (Light Detection And Ranging) is a remote-sensing technique for the measurement of the distance between the sensor and a target. A LiDAR-based detection procedure was tested for characterisation of the weed vegetation present in the inter-row area of a maize field. This procedure was based on the hypothesis that weed species with different heights can be precisely detected and discriminated using non-contact ranging sensors such as LiDAR. The sensor was placed in the front of an all-terrain vehicle, scanning downwards in a vertical plane, perpendicular to the ground, in order to detect the profile of the vegetation (crop and weeds) above the ground. Measurements were taken on a maize field on 3 m wide (0.45 m²) plots at the time of post-emergence herbicide treatments. Four replications were assessed for each of the four major weed species: Sorghum halepense, Cyperus rotundus, Datura ferox and Xanthium strumarium. The sensor readings were correlated with actual, manually determined, height values (r² = 0.88). With canonical discriminant analysis the high capabilities of the system to discriminate tall weeds (S. halepense) from shorter ones were quantified. The classification table showed 77.7% of the original grouped cases (i.e., 4800 sampling units) correctly classified for S. halepense. These results indicate that LiDAR sensors are a promising tool for weed detection and discrimination, presenting significant advantages over other types of non-contact ranging sensors such as a higher sampling resolution and its ability to scan at high sampling rates.     

Predicting drift from field spraying by means of a 3D computational fluid dynamics model

In order to investigate and understand drift from field sprayers, a steady state computational fluid dynamics (CFD) model was developed. The model was developed in 3D in order to increase the understanding of the causes of drift: a deviation in the wind direction cannot be captured by a 2D approach, the wake behind a wind screen is not symmetrical, the effects of a changed nozzle orientation may not be symmetrical. The model's accuracy was validated with field experiments carried out according to the international standard ISO 22866. A field sprayer with a spray boom width of 27 m and 54 nozzles (Hardi ISO F110-03 at 3 bar) was driving at 2.22 m/s over a flat pasture. During the experiments the wind direction was perpendicular to the tractor track. The model explained the variation in drift replicates during each single field experiment through varying boom height (0.3–0.7 m), wind velocity (1.3–2.5 m/s), wind deviation (−18° to +18°) from the direction perpendicular to the tractor track and injection velocity of the droplets (17–27 m/s). Boom movements had the highest impact on the variations in drift values (deviations in drift deposits of 25%), followed by variation in wind velocity (deviations in drift deposits of 3%) and injection velocity of the droplets (deviations in drift deposits of 2.5%). Wind deviation from the direction perpendicular to the tractor track had a reducing effect on the drift values (deviations in drift deposits of 2%). Small variations in driving speed had little influence on drift values. Near drift (<5 m) is predicted well by the model but the increased complexity compromised the predictions at greater distances. The model will be further developed in order to improve far drift prediction. Dynamic simulations will be performed and the model for turbulent dispersion will be optimized. The model did not require calibration.     

Characterization and modeling of water movement and salts transfer in a semi-arid region of Tunisia (Bou Hajla,Kairouan) – Salinization risk of soils and aquifers

The semi-arid region of Bou Hajla (Kairouan – Central Tunisia) is exposed to the risk of soils and aquifers salinization. A characterization of water movement through the use of TDR probes installed at up to 4 m in depth, and of salts transfer by soil sampling was conducted to highlight this risk. At the same time, climatic parameters were monitored. The results were gathered over the period of approximately 1 year (June 12, 2006 to May 08, 2007). Water infiltration and evaporation, occurred at within distinct materials: two zero flux planes at 1 and 3 m, and a clay-silt layer at 2 m where the soil water content and electrical conductivity is highly varied. Saline profiles show salt accumulation at this layer where the electrical conductivity of soil paste extract exceeds 20 dS m⁻¹. However, the additive salt balance between soil surface and a depth of 4 m highlights the salt transfer beyond the studied unsaturated zone, increasing the salinization risk of the aquifer. The water movement and salts transfer were simulated by Hydrus-1d model, and inverse modeling was used to optimize the required hydrodynamic parameters. It was found that the simulated profiles of volumetric humidity and the electrical conductivity of soil are close to those measured. The calculated RMSE values are low, indicating the reliability of Hydrus-1d for the simulation of the hydro-saline dynamics in field conditions.     

A liquid crystal tunable filter based shortwave infrared spectral imaging system: Design and integration

This paper presents the methodology to design and integrate a liquid crystal tunable filter (LCTF) based shortwave infrared (SWIR) spectral imaging system. The system consisted of an LCTF-based SWIR spectral imager, an illumination unit, a frame grabber, and a computer with the data acquisition software. The spectral imager included an InGaAs camera (320 × 256 pixels), an SWIR lens (50 mm, F/1.4), and an LCTF (20 mm aperture). Four multifaceted reflector halogen lamps (35 W, 12 VDC) were used to build the illumination unit. The system was integrated by a LabVIEW program for data acquisition. It can capture hyperspectral or multispectral images of the test object in the spectral range of 900–1700 nm. The system was validated by differentiating sugar from wheat flour, and water from 95% ethanol. The results showed that the system can distinguish these materials in both spectral and spatial domains. This SWIR spectral imaging system could be a potential useful tool for nondestructive inspection of food quality and safety.     

Hyperspectral detection of rice damaged by rice leaf folder (Cnaphalocrocis medinalis)

The reflectance from rice (Oryza sativa L.) leaves and canopy damaged by rice leaf folder (RLF), Cnaphalocrocis medinalis (Guenée) was studied at the booting stage in order to establish a monitoring method for RLF based on hyperspectral data. The results showed that reflectance from rice leaves significantly decreased in the green (530–570 nm) and near infrared (700–1000 nm) regions, and significantly increased in the blue (450–520 nm) and red (580–700 nm) regions as the leaf-roll rate of rice increased. Reflectance from rice canopy significantly decreased in the spectral regions from 737 to 1000 nm as the infestation scale of RLF increased, and the most correlation appeared at 938 nm. Seven spectral regions 503–521, 526–545, 550–568, 581–606, 688–699, 703–715, and 722–770 nm at leaf-level, and one region 747–754 nm at canopy-level were found to be sensitive bands to exhibit the damage severity in rice by RLF. The position of the red edge peak remarkably moved to blue region, and the amplitude and area of the red edge significantly decreased when rice leaves were severely infected by RLF. Thirty-eight spectral indices at leaf-level and 29 indices at canopy-level were found to be sensitive to leaf-roll rate and infestation scale in rice, respectively. The linear regression models were built to detect the leaf-roll rate (0.0–1.0) and infestation scale (0–5) in rice using leaf- and canopy-level reflectance data. The root mean square error of the model was only 0.059 and 0.22 for the leaf-roll rate and infestation scale, respectively. These results suggested that the hyperspectral reflectance was potential to detect RLF damage severity in rice.