Studying the adoption of precision agriculture (PA) is important to identify future challenges and research questions. Currently, there is no mechanism to monitor and compare PA adoption at the European level. This study addressed this issue and proposed a monitoring mechanism based on remote sensing adoption as a case study. A study was set up in six EU countries (France, Germany, Hungary, Italy, Slovakia, and Spain). More than 170 stakeholders were interviewed. While the adoption rates varied between countries and usages, the results confirmed an increasing application of remote sensing in agriculture.
Fruit size information through the season is an important parameter allowing growers to better manage the orchard. In this study, a Python based computer vision algorithm for sizing apples directly on-the-tree was developed. The system was made of a consumer-grade depth camera and tested at two distances and different timings through the season, in a Fuji apple orchard. The system exploited a YOLOv5 detection algorithm and a trigonometric approach based on depth information to size the fruits. Results showed potential field application even if a further system improvement to reduce the sizing error (RMSE <10 mm) need to be achieved.
Turfgrass irrigation consumes a large amount of the scarce freshwater in semi-arid areas like Utah. As much as 50% of this irrigation water is wasted. It has been suggested that determining patterns of spatial variability in soil moisture to modify applications with new valve-in-head sprinkler technology can greatly reduce waste. Variable rate irrigation (VRI) studies in traditional agricultural settings have shown that VRI zones do not stay static temporally and need to be frequently re-determined. Electrical conductivity (ECa) data from Geonics EM38 surveys and data from RGB and Thermal IR drone surveys are less expensive to collect than field survey data. These methods are compared here for their ability to accurately estimate spatial patterns of volumetric water content (VWC) for determining within-season temporally variable turfgrass VRI zones.
The textural information derived from Sentinel-2 image at the R5 soybean stage was assessed by machine learning algorithms to identify soybean yield variability. Two texture measures (entropy and second-moment angular) calculated over the vegetation indices, spectral bands, and RGB, comprised the best models. The 3-class yield map produced reached an accuracy of 0.81 and a Kappa coefficient of 0.73 using the support vector machine algorithm, which was slightly better than K-nearest neighbors. The 4-class map kept the general shape but lost details and performance. The empirical method proposed shows potential and should be tested over other conditions.
In the present contribution, a comparison between cosmic ray neutron sensing (CRNS) and gammaray spectroscopy (GRS) for soil spatial mapping is shown. The experiments have been conducted in a walnut field situated in Bondeno (Ferrara), Northern Italy, characterized by a relatively strong soil texture spatial variability ranging from sandy to clay soil from north to south. Data acquired were compared to ground portable time domain reflectometry (TDR) observations and regional soil classification maps. Results showed good performance in distinguishing the two main soil type zones. For these reasons, the results support the applicability of these methods: (1) to obtain preliminary maps for specific soil sampling designs; (2) to have a qualitative understanding of soil texture and moisture variability of the field; and (3) to divide the field into sectors with similar hydrological properties. Further experiments and analyses should be undertaken, however, to understand the effect of the different spatial footprints of each detector and the best timing for undertaking the survey.
The goal of the new digital agriculture technologies for irrigation efficiency (DATI) project is to design and develop new digital agriculture (DA) technological solutions and procedures for crop and soil monitoring with the purpose of optimizing irrigation management. In a field experiment two crops, melon, and tomato, were monitored and three irrigation treatments were applied. Several narrowband indices from hyperspectral imagery and broadband indices from multispectral were calculated to monitor the impact of the different irrigation treatments. ANOVA showed significant differences between control and different amounts of irrigated water treatments in multispectral and hyperspectral indices. Permutation feature importance technique revealed VARI hyperspectral index as the best index for tomato and OSAVI (optimized soil adjusted vegetation) for melon to discriminate between the different irrigation treatments to flag the respective crops for water stress.
The grapevine disease called ‘Flavescence dorée’ (FD) is actively monitored in Europe as it decreases yield and kills grapevines while being highly contagious. In this study, three methods for its automatic diagnosis from images acquired by proximal sensing (RGB camera) are proposed, compared and discussed. Method A uses a convolutional neural network (CNN) classifier applied on raw images. The two other methods both process in two steps: (1) individual symptom detection using a CNNbased box detector and a deep segmentation algorithm; (2) symptom-based diagnosis based either on a random forest classifier (method B) or on a graph neural network (method C). A 6-fold cross validation was performed on 787 images of vines suffering from FD or from other biotic or abiotic stress factors. Methods B and C reached almost equally good results and outperformed method A: they achieved respectively, in (precision, recall), (0.69, 0.81), (0.87, 0.88) and (0.88, 0.88).
This paper presents an automated method for estimating the trunk cross sectional area of fruit trees. An Intel RealSense 435i was used to capture RGB images and point clouds of individual trunks. To segment the trunk in the image from the background, a Masked-attention Mask Transformer model was adopted. The segmentation results were integrated with the 3D point cloud to estimate trunk widths in 3D. The width estimation was evaluated on three diverse datasets collected from a commercial apple orchard using human measurements as ground truth. With a mean absolute error less than 5%, the method is sufficiently accurate to assist orchard operations.