Precision Agriculture最新文献

Accurate estimation and spatial allocation of economic optimum nitrogen (N) rates (EONR) can support sustainable crop production systems by reducing chemical compounds to be applied to the ground while preserving the optimum yield and profitability Smart Farming (SF) techniques such as historical precision agriculture (PA) machinery data, satellite multispectral imagery, and on-machine nitrogen adjustment sensors can bring together state-of-the-art precision in determining EONR. The novelty of this study is in introducing an efficient optimization framework using SF technology to enable real-time and prescription based EONR application execution. An optimization strategy called response surface modelling (RSM) was implemented to support decision making by fusing multiple sources of information while keeping the underlying computation simple and interpretable. Here, a field of winter wheat with an area of 7 ha was used to prove the proposed concept of determining EONR for each location in the field using auxiliary variables called multispectral indices (MSIs) derived from Sentinel 2. Three different image acquisition dates before the actual N application were considered to find the best time combination of MSIs along with the best MSIs to model yield. The best MSIs were filtered out through three phases of feature selection using analysis of variance (ANOVA), Lasso regression, and model reduction of RSM. For the date 2020.03.25, 14 out of 21 MSIs exhibited a significant interaction with the N applied as determined through an on-machine N sensor. For dates 2020.03.30 and 2020.04.04, the numbers of significant indices were identified as 6 and 10, respectively. Some of the MSIs were no longer significant after five days of the growth period (5-day interval between Sentinel 2 revisits). The best model demonstrated an average prediction error of 14.5%. Utilizing the model’s coefficients, the EONR was computed to be between 43 kg/ha and 75 kg/ha for the target field. By incorporating MSIs into the fitted model for a given N range, it was demonstrated that the shape of the yield-N relation (RSM) varied due to field heterogeneity. The proposed analytical approach integrates farmer engagement by participatory annual post-mortem analysis. Using the determined RSM approach, retrospective assessment compares economically optimal N input, based on observed MSIs values to each location, with the actual applied rates.

The ability to reduce sampling distance or time is crucial for growers to monitor vineyard nutrients more frequently. Extension specialists often recommend collecting large random samples, but this is frequently overlooked, leading to inaccurate fertilizer recommendations. A novel, one-location square grid area-based sampling method called “box” sampling was developed to capture the overall nutrient distribution within a block, providing guidance for growers on sample collection in vineyards for nutrient monitoring. Box sampling was compared with random and stratified sampling methods at both bloom and veraison for grapevine foliar nitrogen (N%), phosphorus (P%), potassium (K%), magnesium (Mg%), and calcium (Ca%). Box and stratified sampling locations were determined based on Synthetic Aperture Radar (SAR) from Sentinel-1 and Sentinel-2 Normalized Difference Vegetation Index (NDVI) images. SAR and NDVI images were stratified into three variability zones using the k-means + + algorithm. Representative pixels from each zone were sampled using the stratified method, while the junction of these variability zones (30mx30m sampling window) was sampled using the new box method. In 2021 and 2022, these methods were compared against nutrient population parameters in two vineyard blocks. Both methods showed marginal differences in mean, median, and standard deviation, with box sampling consistently capturing a broader range of variations. This was evidenced by the Bhattacharya coefficient, which indicates the overlap between two probability distributions (with values closer to 1 for greater overlap). The coefficient was > 0.80 for N%, P%, and Mg%, and > 0.60 for K% and Ca% at both bloom and veraison. For 14 different commercial vineyards in 2022 and 2023, box sampling accurately captured random nutrient variability for N%, P% and Mg% at both bloom and veraison. However, for K% (at veraison) and Ca% box sampling performed poorly due to high spatial variability. Box sampling reduced the sampling distance and time by 75% compared to random sampling.

Evaluated Sentinel-2, SuperDove, and Dove-R consistency for wheat field monitoring.

Hierarchical evaluation on surface reflectance, VIs, and LAI.

VI and LAI consistencies of Sentinel-2 and PlanetScope exceed surface reflectance.

Sentinel-2 and PlanetScope’s optimal synergy interval at VI or LAI is 2 days.

Nitrogen (N) is a key factor affecting corn yield. Remote sensing of spectral reflectance from plant canopies offers an efficient way to assess N status. High spatial and temporal resolution imagery from unmanned aerial vehicles (UAVs) provides additional advantages. This study aimed to (1) develop and validate a model to predict top-dressing N requirements at the V5 stage using vegetation indices (VIs), N rates, and/or leaf N content (LNC), and (2) correlate VIs with LNC and yield at V6, V11, and R1 stages. Two experiments were conducted in Goiás state, Brazil. The first tested N rates from 0 to 300 kg ha⁻¹ applied at V5, with imagery and LNC collected at V6, V11, and R1 stages. VIs such as GNDVI (R² = 0.55–0.74), GN (R² = 0.70–0.75), and TCARI (R² = 0.62–0.63) showed strong correlations with N sources and LNC. Linear, linear-plateau, and quadratic-plateau models best fit the data. The validation trial confirmed the effectiveness of these VIs in optimizing N applications without reducing yield. GNDVI presented more benefits of reducing the amount of top-dressed N regardless of the variable used (N rate or LNC). The reduction of N inputs ranged from 6.6 to 35% compared to traditional methods. Additionally, VIs such as SAVI, GSAVI, and RVI accurately predicted yield, especially at the V6 stage, where correlations were highest (R² ≥ 0.70). This approach demonstrates the potential of UAV-based VIs for optimizing N management and improving grain yield predictions.

Mechanical weeding is an important part of integrated weed management. It destroys weeds between (interrow) and in (intrarow) crop rows. Preventing crop damage requires precise detection and tracking of the plants. In this work, a detection and tracking algorithm was developed and integrated on an intrarow hoeing prototype. The algorithm was developed and validated on 12 rows of 950 cauliflower plants. Therefore, a methodology was provided to automatically generate a label based on the crop plants’ Global Navigation Satellite System (GNSS) position during data collection with a robot platform. A CenterNet architecture was adjusted for plant centre detection by comparing different encoder networks and selecting the optimal hyperparameters. The monocular camera projection error of the plant centre detections in pixel to 3D coordinates was evaluated and used in a position- and velocity-based tracking algorithm to determine the timing for intrarow hoeing knife actuation. A dataset of 53k labelled images was created. The best CenterNet model resulted in an F1 score on the test set of 0.986 for detecting cauliflower centres. The position tracking had an average variation of 1.62 cm. Velocity tracking had a standard deviation of 0.008 (mathrm {m,,s^{-1}}) with respect to the robot’s operational target velocity. Overall, the entire integration showed effective actuation of the prototype in field conditions. Only one false positive detection occurred during operation in two test rows of 135 cauliflowers.

Optimizing the timing of rice paddy drainage and harvest is crucial for maximizing yield and quality. These decisions are guided by rice grain moisture content (GMC), which is typically determined by destructive plant samples taken at point locations. Providing rice farmers with predictions of GMC will reduce the time burden of gathering, threshing and testing samples. Additionally, it will reduce errors due to samples being taken from unrepresentative areas of fields, and will facilitate advanced planning of end-of-season drain and harvest timing. This work demonstrates consistent relationships between rice GMC and indices derived from Sentinel-2 satellite imagery, particularly those involving selected shortwave infrared and red edge bands (r=0.84, 1620 field samples, 3 years). A methodology was developed to allow forecasts of grain moisture past the latest image date to be provided, by fusing remote sensing and accumulated weather data as inputs to machine learning models. The moisture content predictions had root mean squared error between 1.6 and 2.6% and (hbox {R}^2) of 0.7 with forecast horizons from 0 to 28 days. Time-series grain moisture dry-down predictions were summarized per field to find the optimal harvest date (22% grain moisture), with an average RMSE around 6.5 days. The developed methodology was operationalized to provide rice growers with current and projected grain moisture, enabling data-driven decisions, ultimately enhancing operational efficiency and crop outcomes.

Purpose

Extracting the extent of wheat lodging is essential for post-disaster emergency response, disaster assessment, and accurate agricultural insurance claims. However, traditional methods for identifying lodged crops often lack flexibility, exhibit low levels of automation, and suffer from inefficiency.

Methods

This study proposes a rapid identification algorithm for wheat lodging, utilizing adaptive thresholding and a two-peak search of UAV imagery for reliable extraction of lodging regions. Initially, the red, green, and blue (RGB) visible band characteristics of UAV images after wheat lodging are analyzed. Subsequently, an Enhanced Wheat Lodging Index (EWLI) is proposed to quantitatively represent the lodging state. Second, a two-peak search dynamic thresholding algorithm, based on the square chunking of wheat lodging, is proposed to automatically determine thresholds for extracting winter wheat lodging regions.

Results

Experimental results demonstrate that the Enhanced Wheat Lodging Index (EWLI) effectively represents wheat lodging, while the two-peak search dynamic thresholding algorithm achieves robust performance. The proposed method achieves an overall accuracy of 96%, an F1 score of 0.97, and a Kappa coefficient exceeding 0.95, surpassing the performance of the OTSU method (maximum inter-class variance) and the KSW method (maximum entropy) with global thresholding.

Conclusion

The proposed method is applicable to diverse wheat lodging scenarios and demonstrates robust stability in identification accuracy. Key advantages include lightweight modeling, adaptive threshold determination, and the elimination of human intervention, making it an efficient, reliable, and highly practical approach for wheat lodging monitoring.

The use of thermal-based crop water stress index (CWSI) has been studied in many crops in semi-arid regions and found as an effective method in detecting real-time crop water status of commercial fields remotely and non-destructively. However, to our knowledge, no previous studies have validated the usefulness of CWSI in a temperate crop like wild blueberries. Additionally, the temporal changes of the water status estimation model has not been well-studied. In this multi-year study, Unoccupied Aerial Vehicle (UAV)-borne thermal imageries were collected in 2019, 2020, and 2021 to test the temporal effects and the impact of different approach-based reference temperatures (T_wet, wet reference temperature; T_dry, dry reference temperature) on leaf water potential (LWP) estimation models using CWSI in two large adjacent wild blueberry fields in Maine, United States. We found that different sampling dates have a significant impact on LWP estimation models using CWSI_SE (statistical T_wet and empirical T_dry reference) and CWSI_SS (statistical T_wet and statistical T_dry reference). Further, CWSI_BB calculated with bio-indicator-based T_wet and T_dry reference was found more effective (r² = 0.79) in estimating LWP in 2021, compared to the CWSI_SE and CWSI_SS approaches in 2019 (r² = 0.34 & r² = 0.36), 2020 (r² = 0.38 & r² = 0.44) and 2021 (r² = 0.43 & r² = 0.46). CWSI_BB -LWP model-based crop water status maps show high variation in the crop water status of wild blueberries, even in an evenly irrigated field, suggesting the potential of UAV-borne thermal cameras to detect real-time crop water status within the field, with the CWSI_BB calculated from bio-indicator-based references being more reliable. Our results could be used for precision irrigation to increase the overall water use efficiency and profitability of wild blueberry production.

Durum wheat (Triticum durum Desf.) yield should be maximized to meet the growing global demand for pasta production. Precision agriculture (PA) could play a pivotal role in reaching this goal by correctly defining management zones (MZ) and optimizing the use of energy inputs. The aim of the work was to understand the relationship between MZ generated from observed yield data and those generated using a time series of Sentinel-derived vegetation indices (i.e. NDVI) obtained from satellite images and soil properties. For this purpose, two field trials of 10 ha each, cultivated with durum wheat, were carried out in Southern Italy. The results suggested a better strategy for defining MZs by merging soil characteristics and temporal NDVI stability maps. The on-the-go technology used for soil resistivity mapping also represented an excellent tool for delineating stable and homogeneous areas within the fields and estimating soil properties. In particular, the soil clay content had a determining effect on the identification of homogeneous yield areas. However, the integration of historical NDVI data helped delineate MZs within each field. To validate this hypothesis, we integrated soil and NDVI data into a linear predictive model to predict grain yield at the field level. Our findings showed a good level of accuracy and a significant improvement in yield simulated values by combining soil with crop data (R² = 0.620; RMSE = 0.425). Further studies are needed to explore the potential of NDVI stability maps into a linear predictive model to predict grain yield at the field level.

Context

Serrated tussock (Nassella trichotoma) is a weed of national significance in Australia which offers little to no nutritional value to livestock, and has the potential to reduce carrying capacity and agricultural return of infested pastures.

Aims

The aim of this study was to adapt existing Convolutional Neural Networks (CNNs) for plant segmentation and spraypoint detection in the challenging environments of pastures.

Methods

CNNs that were designed for joint plant and stem segmentation in crop fields were repurposed for dual-task applications in pastures. Given the poor performance of these models in complex pasture environments, a new model drawing inspiration from the recently proposed ConvNeXt was developed, tested for its effectiveness on unseen field data, and enhanced with a novel normalization technique, called HistMatch.

Key results

Experimentation demonstrated that unlike pre-existing models, which were designed for the simpler environments encountered in early-stage crop fields, our model was able to generalize well to growing conditions not seen during training, achieving 0.807 mIoU and 0.796 F1-score for the plant and spraypoint tasks respectively. This is in comparison to pre-existing models, which achieved 0.270 - 0.454 mIoU and 0.073 - 0.496 F1-score for the same tasks. These results were further improved to 0.854 mIoU and 0.806 F1-score using HistMatch normalization. In spite of greater model complexity, our model had a inference time of 15.7 ms which was comparable to pre-existing models, and suitable for real-time applications.

Conclusion

Models with greater complexity are required for the relatively complex environments encountered in pastures, but this greater complexity need not come at the expense of real time capability. HistMatch normalization can improve model accuracy, and is particularly effective in cases where models are struggling to generalize well to testing conditions that vary significantly from those seen during training.

Implications and impacts

The successful adaptation and improvement of CNNs for weed management in pastures could significantly reduce the reliance on blanket herbicide application. HistMatch normalization could also be considered for other agricultural applications, including weed management and disease detection in crop fields and orchards.