Nigel Hancock

Applied machine vision of plants: a review with implications for field deployment in automated farming operations

Automated visual assessment of plant condition, specifically foliage wilting, reflectance and growth parameters, using machine vision has potential use as input for real-time variable-rate irrigation and fertigation systems in precision agriculture. This paper reviews the research literature for both outdoor and indoor applications of machine vision of plants, which reveals that different environments necessitate varying levels of complexity in both apparatus and nature of plant measurement which can be achieved. Deployment of systems to the field environment in precision agriculture applications presents the challenge of overcoming image variation caused by the diurnal and seasonal variation of sunlight. From the literature reviewed, it is argued that augmenting a monocular RGB vision system with additional sensing techniques potentially reduces image analysis complexity while enhancing system robustness to environmental variables. Therefore, machine vision systems with a foundation in optical and lighting design may potentially expedite the transition from laboratory and research prototype to robust field tool.

Advanced process control of irrigation: the current state and an analysis to aid future development

Control engineering approaches may be applied to irrigation management to make better use of available irrigation water. These methods of irrigation decision-making are being developed to deal with spatial and temporal variability in field properties, data availability and hardware constraints. One type of control system is advanced process control which, in an irrigation context, refers to the incorporation of multiple aspects of optimisation and control. Hence, advanced process control is particularly suited to the management of site-specific irrigation. This paper reviews applications of advanced process control in irrigation: mathematical programming, linear quadratic control, artificial intelligence, iterative learning control and model predictive control. From the literature review, it is argued that model-based control strategies are more realistic in the soil–plant–atmosphere system using process simulation models rather than using ‘black-box’ crop production models. It is also argued that model-based control strategies can aim for specific end of season characteristics and hence may achieve optimality. Three control systems are identified that are robust to data gaps and deficiencies and account for spatial and temporal variability in field characteristics, namely iterative learning control, iterative hill climbing control and model predictive control: from consideration of these three systems it is concluded that the most appropriate control strategy depends on factors including sensor data availability and grower’s specific performance requirements. It is further argued that control strategy development will be driven by the available sensor technology and irrigation hardware, but also that control strategy options should also drive future plant and soil moisture sensor development.

Managing Spatial and Temporal Variability in Irrigated Agriculture through Adaptive Control

Abstract Spatial variability in crop production occurs as a result of spatial and temporal variations in soil structure and fertility; soil physical, chemical and hydraulic properties; irrigation applications; pests and diseases; plant genetics; and local microclimate. This review paper argues that infield variability can be managed and the efficiency of irrigation water use increased by spatially variable application of irrigation water to meet the specific needs of individual management zones (areas of crop whose properties are relatively homogenous). Key areas identified requiring interdisciplinary research are the prescription of irrigated crop water requirements, strategies for quantifying and managing spatial variability, and the development of adaptive systems for control of water application at appropriate temporal intervals and spatial scales. Example strategies for the implementation of adaptive control for furrow irrigation and large mobile irrigation machines are described.

Orchard mapping and mobile robot localisation using on-board camera and laser scanner data fusion - Part A

A novel tree trunk detection algorithm uses camera and laser scanner data fusion.Discriminates between trees and non-tree objects in the orchard.Automatic adjustment of colour parameters increased the detection accuracy.Evaluation under different illumination conditions (sunny, cloudy).Using a small robot platform, a detection accuracy of 96.64% was achieved in a real orchard. Trees in orchards are natural landmarks providing suitable cues for mobile robot localisation as they are nominally planted in straight and parallel rows. This paper presents a novel tree trunk detection algorithm using a camera and laser scanner data fusion to enhance the detection capability. The algorithm detects the trees in the orchard and discriminates between trees and non-tree objects (e.g. posts and tree supports). The laser scanner is used to detect the edge points and determine the width of the tree trunks and non-tree objects, while the camera images are used to verify the colour and the parallel edges of the tree trunks and non-tree objects. The algorithm automatically adjusts the colour detection parameters after each test which shown to increase the detection accuracy. Experimental tests were conducted with a small robot platform in a real orchard environment to evaluate the performance of the tree trunk detection algorithm under two broad illumination conditions (sunny and cloudy). The algorithm was able to detect the tree trunks and discriminate between trees and non-tree objects with detection accuracy of 96.64% showing that the fusion of both vision and laser scanner technologies produced robust tree trunk detection.

Automated Internode Length Measurement of Cotton Plants under Field Conditions

An in-field vision system has been developed that automatically and non-destructively measures internode length (i.e., the distance between successive main stem branches) of plants in a growing cotton crop for the purpose of inferring plant growth performance and informing crop irrigation management. The system uses monocular video acquisition behind a plant-contacting transparent panel that moves through the crop. Line features are extracted from acquired imagery to estimate candidate nodes on the plants main stem via a two-stage process that may be implemented in real-time. Firstly, candidate nodes are identified in each individual image; secondly, the comparison of sequential images permits the removal of erroneously identified nodes (false positives) and compensation for missed nodes (due to occlusion by a leaf, commonly). Ninety-five internode length measurements were automatically detected from 168 video sequences of 14 plants. Algorithm run-time calculations and the rate of internode measurement acquisition were shown to be adequate for real-time input to spatially varied irrigation control. For this data set, the median absolute error was 5.3 mm in comparison with physical plant measurements, and the standard error in measurement ranged from 1.1 to 5.7 mm, with an average of 3.0 mm.

Orchard mapping and mobile robot localisation using on-board camera and laser scanner data fusion – Part B: Mapping and localisation

Accurate mobile robot localisation in orchards relies on precise orchard maps which help the mobile robot to efficiently estimate its position and orientation while moving between tree rows. This paper presents a new method for constructing a local orchard map based on tree trunk detection using camera and laser scanner data fusion. The final orchard map consists of the positions of the trees and non-tree objects (e.g. posts and tree supports) in the tree rows. The map of the individual trees is used as an a priori map to localise the mobile robot in the orchard. A data fusion algorithm based on an Extended Kalman Filter is used for position estimation. Experimental tests were conducted with a small robot platform in a real orchard environment to evaluate the performance of orchard mapping and mobile robot localisation. The mapping method successfully localised all the trees and non-tree objects of the tested tree rows in the orchard. The mapping results indicate that the constructed orchard map can be reliably used for mobile robot localisation and navigation. The localisation algorithm was evaluated against the logged RTK-GPS positions for different paths and headland turns. The average of the root mean square of the Euclidean distance between the ground truth and the estimated position for different paths was 0.103 m, whilst the average of the root mean square of the heading error was 3.32°.

Real-time data requirements for model-based adaptive control of irrigation scheduling in cotton

Abstract Model-based adaptive control strategies can be used to determine site-specific irrigation volumes with the aim of maximising crop water use efficiencies and/or yield. These strategies require infield weather, soil and crop measurements to calibrate a crop model: the crop model is then used to determine the irrigation applications throughout the crop season which produce the desired simulated crop response or condition (eg. maximum yield). However, data collection spatially over a field and throughout the crop season will potentially lead to a large sensed data requirement which may be impractical in a field implementation. Not all the collected data may be required to sufficiently calibrate the crop model and determine irrigation applications for model-based adaptive control; rather, a smaller dataset consisting of only the most influential sensor variables may be sufficient for adaptive control purposes. This paper reports on afield study which examined the utility of five sensed variables – evaporative demand, soil moisture, plant height, square count and boll count – to calibrate the cotton model OZCOT within a model-based controller and evaluate the relative significance of each sensed variable (either individually or in combination) as a control input. For the field study conditions, OZCOT was most effectively calibrated (and therefore able to predict the soil and crop response to irrigation application) using full data input, while for situations where only two data inputs were available, the simulations suggested that either weather-and-plant or soil-and-plant inputs were preferable.

On-the-go machine vision sensing of cotton plant geometric parameters: first results

Plant geometrical parameters such as internode length (i.e. the distance between successive branches on the main stem) indicate water stress in cotton. This paper describes a machine vision system that has been designed to measure internode length for the purpose of determining real-time cotton plant irrigation requirement. The imaging system features an enclosure which continuously traverses the crop canopy and forces the flexible upper main stem of individual plants against a glass panel at the front of the enclosure, hence allowing images of the plant to be captured in a fixed object plane. Subsequent image processing of selected video sequences enabled detection of the main stem in 88% of frames. However, node detection was subject to a high false detection rate due to leaf edges present in the images. Manual identification of nodes in the acquired imagery enabled measurement of internode lengths with 3% standard error.

Development of a smart monolayer application system for reducing evaporation from farm dams: introductory paper

Abstract Chemical monolayer films are potentially an economical low-impact means of reducing evaporative loss from farm water storages. However, their performance can be highly variable as they are affected by climatic and environmental factors: principally wind, wave action and bio-degradation. Some of this observed variability is associated with the monolayer materials themselves and their interaction with the water-surface physics and biology, but the fact that they are only a few nanometres thick means that a very small amount of material has to be distributed over a very large area. Therefore, appropriate and timely autonomous application of monolayer, with regard to prevailing (and changing) wind conditions on-site, is required. Although a number of autonomous application systems for monolayer already exist, none has proved overly successful. It is argued that while this is in part due to sub-optimal performance of monolayer materials, it is also due in large measure to inaccuracies and/or inappropriate design in both application systems and particularly application strategies, which are not adaptive to the prevailing environmental conditions. Therefore a control system is being developed to adaptively and spatially vary monolayer application rates according to changing conditions monitored on-site. This will form part of an autonomous electromechanical system for the optimal application and spreading of any given chemical monolayer. This paper reports progress towards this objective; firstly by evaluation of the design requirements for automated systems at a range of spatial scales; and secondly via the construction of a first pre-prototype to act as an evaluation platform and concept demonstrator.

Non-contact extrudate profilometer - introductory paper

A non-contact measurement system has been devised to monitor the cross-section of newly-formed plastic product (extrudate) during mechanical extrusion processes. A laser source is reflected from the surface and the light beam detected such that the position and orientation of a point on the surface is determined. This paper describes the initial development of the profilometer extrudate of square cross-section. Preliminary results are reported which indicate that measurement precision of better than 10 /spl mu/m is achievable. Developments necessary to facilitate circumferential scanning of the extrudate surface are discussed.