No abstract

No abstract

Plant breeding is an extremely important route to genetic improvements that can increase yield and plant adaptability. Genetic improvement requires careful measurement of plant phenotypes or plant trait characteristics, but phenotype measurement is a tedious and error-prone task for humans to perform. Highthroughput phenotyping aims to eliminate the problems of manual phenotype measurement. In this paper, we propose and demonstrate the efficacy of an automatic corn plant phenotyping system based on 3D holographic reconstruction. Point cloud image data were acquired from a time-of-flight 3D camera, which was integrated with a plant rotating table to form a screening station. Our method has five main steps: point cloud data filtering and merging, stem segmentation, leaf segmentation, phenotypic data extraction, and 3D holographic visualization. In an experimental study with five corn plants at their early growth stage (V3), we obtained promising results with accurate 3D holographic reconstruction. The average measurement error rate for stem major axis, stem minor axis, stem height, leaf area, leaf length and leaf angle were at 7.92%, 15.20%, 7.45%, 21.89%, 10.25% and 11.09%, respectively. The most challenging trait to measure was leaf area due to partial occlusions and rolling of some leaves. In future work, we plan to extend and evaluate the usability of the system in an industrial plant breeding setting.

Field operations should be done in a manner that minimizes time and travels over the field surface and is coordinated with topographic land features. Automated path planning can help to find the best coverage path so that the field operation costs can be minimized. Intelligent algorithms are desired for both two-dimensional (2D) and three-dimensional (3D) terrain field coverage path planning. The algorithm of generating an optimized full coverage pattern for a given 2D planar field by using boustrophedon paths has been investigated and reported before. However, a great proportion of farms have rolling terrains, which have a considerable influence on the design of coverage paths. Coverage path planning in 3D space has a great potential to further optimize field operations. This work addressed four critical tasks: terrain modeling and representation, coverage cost analysis, terrain decomposition, and the development of optimized path searching algorithm. The developed algorithms and methods have been successfully implemented and tested using 3D terrain maps of farm fields with various topographic features. Each field was decomposed into subregions based on its terrain features. A recommended "seed curve" based on a customized cost function was searched for each subregion, and parallel coverage paths were generated by offsetting the found "seed curve" toward its two sides until the whole region was completely covered. Compared with the 2D planning results, the experimental results of 3D coverage path planning showed its superiority in reducing both headland turning cost and soil erosion cost. On the tested fields, on average the 3D planning algorithm saved 10.3% on headland turning cost, 24.7% on soil erosion cost, 81.2% on skipped area cost, and 22.0% on the weighted sum of these costs, where their corresponding weights were 1, 1, and 0.5, respectively. C 2011 Wiley Periodicals, Inc.

Though some two-dimensional (2D) machine vision-based systems for early-growthstage corn plant sensing exist, some of their shortcomings are difficult to overcome. The greatest challenge comes from separating individual corn plants with overlapped plant canopies. With 2D machine vision, variation in outdoor lighting conditions and weeds in the background also pose difficulties in corn plant identification. Adding the depth dimension has the potential to improve the performance of such a sensing system. A new corn plant sensing system using a real-time stereo vision system was investigated in this research. Top-view depth images of corn plant canopy were acquired. By processing the depth images, the algorithm effectively updated the plant skeleton structures and finally recognized individual corn plants and detected their center positions. The stereo vision system was tested over corn plants of V2-V3 growth stages in both laboratory and field conditions. Experimental results showed that the stereo vision system was capable of detecting both separated and overlapped corn plants. During the field test, 96.7% of the corn plants were correctly detected, and plant center positions were estimated with maximum distance errors of 5 and 1 cm for 74.6% and 62.3% of detections, respectively. C 2009 Wiley Periodicals, Inc.

A texture-based weed classification method was developed. The method consisted of a low-level Gabor wavelets-based feature extraction algorithm and a high-level neural network-based pattern recognition algorithm. This classification method was specifically developed to explore the feasibility of classifying weed images into broadleaf and grass categories for spatially selective weed control. In this research, three species of broadleaf weeds (common cocklebur, velvetleaf, and ivyleaf morning glory) and two grasses (giant foxtail and crabgrass) that are common in Illinois were studied. After processing 40 sample images with 20 samples from each class, the results showed that the method was capable of classifying all the samples correctly with high computational efficiency, demonstrating its potential for practical implementation under real-time constraints.

No abstract

No abstract