@article{lu_young_linder_whipker_suchoff_2022, title={Hyperspectral Imaging With Machine Learning to Differentiate Cultivars, Growth Stages, Flowers, and Leaves of Industrial Hemp (Cannabis sativa L.)}, volume={12}, ISSN={["1664-462X"]}, DOI={10.3389/fpls.2021.810113}, abstractNote={As an emerging cash crop, industrial hemp (Cannabis sativa L.) grown for cannabidiol (CBD) has spurred a surge of interest in the United States. Cultivar selection and harvest timing are important to produce CBD hemp profitably and avoid economic loss resulting from the tetrahydrocannabinol (THC) concentration in the crop exceeding regulatory limits. Hence there is a need for differentiating CBD hemp cultivars and growth stages to aid in cultivar and genotype selection and optimization of harvest timing. Current methods that rely on visual assessment of plant phenotypes and chemical procedures are limited because of its subjective and destructive nature. In this study, hyperspectral imaging was proposed as a novel, objective, and non-destructive method for differentiating hemp cultivars, growth stages as well as plant organs (leaves and flowers). Five cultivars of CBD hemp were grown greenhouse conditions and leaves and flowers were sampled at five growth stages 2–10 weeks in 2-week intervals after flower initiation and scanned by a benchtop hyperspectral imaging system in the spectral range of 400–1000 nm. The acquired images were subjected to image processing procedures to extract the spectra of hemp samples. The spectral profiles and scatter plots of principal component analysis of the spectral data revealed a certain degree of separation between hemp cultivars, growth stages, and plant organs. Machine learning based on regularized linear discriminant analysis achieved the accuracy of up to 99.6% in differentiating the five hemp cultivars. Plant organ and growth stage need to be factored into model development for hemp cultivar classification. The classification models achieved 100% accuracy in differentiating the five growth stages and two plant organs. This study demonstrates the effectiveness of hyperspectral imaging for differentiating cultivars, growth stages and plant organs of CBD hemp, which is a potentially useful tool for growers and breeders of CBD hemp.}, journal={FRONTIERS IN PLANT SCIENCE}, author={Lu, Yuzhen and Young, Sierra and Linder, Eric and Whipker, Brian and Suchoff, David}, year={2022}, month={Feb} }
 @article{lu_li_young_li_linder_suchoff_2022, title={Hyperspectral imaging with chemometrics for non-destructive determination of cannabinoids in floral and leaf materials of industrial hemp (Cannabis sativa L.)}, volume={202}, ISSN={["1872-7107"]}, DOI={10.1016/j.compag.2022.107387}, abstractNote={With the passage of the 2018 Farm Bill, industrial hemp (Cannabis sativa L.) has become a legal and economically promising crop commodity for U.S. farmers. There has been a surge of interest in growing industrial hemp for producing cannabinoids, such as cannabidiol (CBD), because of their medical potential. Quantitative determination of cannabinoids in harvested materials (primarily floral tissues) is critical for cannabinoid production and compliance testing. The concentrations of cannabinoids in hemp materials are conventionally determined using wet-chemistry chromatographic methods, which require destructive sampling, and are time-consuming, costly, and thus not suitable for on-site rapid testing. This study presents a novel effort to utilize hyperspectral imaging technology for non-destructive quantification of major cannabinoids, including CBD, THC (tetrahydrocannabinol), CBG (cannabigerol) and their acid forms in fresh floral and leaf materials of industrial hemp on a dry weight basis. Hyperspectral images in the wavelength range of 400–1000 nm were acquired from floral and leaf tissues immediately after harvest from a total of 100 industrial hemp plants of five cultivars at varied growth stages. Linear discriminant analysis showed hyperspectral imaging could identify CBD-rich/poor and THC-legal/illegal flower samples with accuracies of 99% and 97%, respectively. Quantitative models based on full-spectrum PLS (partial least squares) achieved prediction accuracies of RPD (ratio of prediction to deviation) = 2.5 (corresponding R2 = 0.84) for CBD and THC in floral tissues. Similar accuracies were obtained for their acid forms in flower samples. The predictions for CBG and its acid form in floral tissues and all six cannabinoids in leaf tissues were unsatisfactory with noticeably lower RPD values. Consistently improved accuracies were obtained by parsimonious PLS models based on a wavelength selection procedure for minimized variable collinearity. The best RPD values of approximately 2.6 (corresponding R2 = 0.85) were obtained for CBD and THC in floral materials. This study demonstrates the utility of hyperspectral imaging as a potential valuable tool for rapid quantification of cannabinoids in industrial hemp.}, journal={COMPUTERS AND ELECTRONICS IN AGRICULTURE}, author={Lu, Yuzhen and Li, Xu and Young, Sierra and Li, Xin and Linder, Eric and Suchoff, David}, year={2022}, month={Nov} }
 @article{lu_young_wang_wijewardane_2022, title={Robust plant segmentation of color images based on image contrast optimization}, volume={193}, ISSN={["1872-7107"]}, DOI={10.1016/j.compag.2022.106711}, abstractNote={• A contrast-optimization approach was proposed for plant segmentation of color images. • Contrast-enhanced images were compared with index images using five image datasets. • The proposed method consistently enhanced image contrast and segmentation accuracy. • None of nine common color indices were robust enough to varying image conditions. Plant segmentation is a crucial task in computer vision applications for identification/classification and quantification of plant phenotypic features. Robust segmentation of plants is challenged by a variety of factors such as unstructured background, variable illumination, biological variations, and weak plant-background contrast. Existing color indices that are empirically developed in specific applications may not adapt robustly to varying imaging conditions. This study proposes a new method for robust, automatic segmentation of plants from background in color (red-green-blue, RGB) images. This method consists of unconstrained optimization of a linear combination of RGB component images to enhance the contrast between plant and background regions, followed by automatic thresholding of the contrast-enhanced images ( CEI s). The validity of this method was demonstrated using five plant image datasets acquired under different field or indoor conditions, with a total of 329 color images as well as ground-truth plant masks. The CEI s along with 10 common index images were evaluated in terms of image contrast and plant segmentation accuracy. The CEI s, based on the maximized foreground-background separability, achieved consistent, substantial improvements in image contrast over the index images, with an average segmentation accuracy of F1 = 95%, which is 4% better than the best accuracy obtained by the indices. The index images were found sensitive to imaging conditions and none of them performed robustly across the datasets. The proposed method is straightforward, easy to implement and can be potentially extended to nonlinear forms of color component combinations or other color spaces and generally useful in plant image analysis for precision agriculture and plant phenotyping.}, journal={COMPUTERS AND ELECTRONICS IN AGRICULTURE}, author={Lu, Yuzhen and Young, Sierra and Wang, Haifeng and Wijewardane, Nuwan}, year={2022}, month={Feb} }
 @article{lu_payn_pandey_acosta_heine_walker_young_2021, title={HYPERSPECTRAL IMAGING WITH COST-SENSITIVE LEARNING FOR HIGH-THROUGHPUT SCREENING OF LOBLOLLY PINE (PINUS TAEDA L.) SEEDLINGS FOR FREEZE TOLERANCE}, volume={64}, ISSN={["2151-0040"]}, url={http://dx.doi.org/10.13031/trans.14708}, DOI={10.13031/trans.14708}, abstractNote={HighlightsA hyperspectral imaging approach was developed for freeze-tolerance phenotyping of loblolly pine seedlings.Image acquisition was conducted before and periodically after artificial freezing of the seedlings.A hyperspectral data processing pipeline was developed to extract the spectra from seedling segments.Cost-sensitive support vector machine (SVM) was used for classifying stressed and healthy seedlings.Post-freeze scanning of seedlings on day 41 achieved the highest screening accuracy of 97%.Abstract. Loblolly pine (Pinus taeda L.) is a commercially important timber species planted across a wide temperature gradient in the southeastern U.S. It is critical to ensure that the planting stock is suitably adapted to the growing environment to achieve high productivity and survival. Long-term field studies, although considered the most reliable method for assessing cold hardiness of loblolly pine, are extremely resource-intensive and time-consuming. The development of a high-throughput screening tool to characterize and classify freeze tolerance among different genetic entries of seedlings will facilitate accurate deployment of highly productive and well-adapted families across the landscape. This study presents a novel approach using hyperspectral imaging to screen loblolly pine seedlings for freeze tolerance. A diverse population of 1549 seedlings raised in a nursery were subjected to an artificial mid-winter freeze using a freeze chamber. A custom-assembled hyperspectral imaging system was used for in-situ scanning of the seedlings before and periodically after the freeze event, followed by visual scoring of the frozen seedlings. A hyperspectral data processing pipeline was developed to segment individual seedlings and extract the spectral data. Examination of the spectral features of the seedlings revealed reductions in chlorophylls and water concentrations in the freeze-susceptible plants. Because the majority of seedlings were freeze-stressed, leading to severe class imbalance in the hyperspectral data, a cost-sensitive learning technique that aims to optimize a class-specific cost matrix in classification schemes was proposed for modeling the imbalanced hyperspectral data, classifying the seedlings into healthy and freeze-stressed phenotypes. Cost optimization was effective for boosting the classification accuracy compared to regular modeling that assigns equal costs to individual classes. Full-spectrum, cost-optimized support vector machine (SVM) models achieved geometric classification accuracies of 75% to 78% before and within 10 days after the freeze event, and up to 96% for seedlings 41 days after the freeze event. The top portions of seedlings were more indicative of freeze events than the middle and bottom portions, leading to better classification accuracies. Further, variable selection enabled significant reductions in wavelengths while achieving even better accuracies of up to 97% than full-spectrum SVM modeling. This study demonstrates that hyperspectral imaging can provide tree breeders with a valuable tool for improved efficiency and objectivity in the characterization and screening of freeze tolerance for loblolly pine. Keywords: Cost-sensitive learning, Freeze tolerance, Hyperspectral imaging, Plant phenotyping, Support vector machine.}, number={6}, journal={TRANSACTIONS OF THE ASABE}, publisher={American Society of Agricultural and Biological Engineers (ASABE)}, author={Lu, Yuzhen and Payn, Kitt G. and Pandey, Piyush and Acosta, Juan J. and Heine, Austin J. and Walker, Trevor D. and Young, Sierra}, year={2021}, pages={2045–2059} }
 @article{lu_walker_acosta_young_pandey_heine_payn_2021, title={Prediction of Freeze Damage and Minimum Winter Temperature of the Seed Source of Loblolly Pine Seedlings Using Hyperspectral Imaging}, volume={67}, ISSN={["1938-3738"]}, url={https://doi.org/10.1093/forsci/fxab003}, DOI={10.1093/forsci/fxab003}, abstractNote={
 The most important climatic variable influencing growth and survival of loblolly pine is the yearly average minimum winter temperature (MWT) at the seed source origin, and it is used to guide the transfer of improved seed lots throughout the species’ distribution. This study presents a novel approach for the assessment of freeze-induced damage and prediction of MWT at seed source origin of loblolly pine seedlings using hyperspectral imaging. A population comprising 98 seed lots representing a wide range of MWT at seed source origin was subjected to an artificial freeze event. The visual assessment of freeze damage and MWT were evaluated at the family level and modeled with hyperspectral image data combined with chemometric techniques. Hyperspectral scanning of the seedlings was conducted prior to the freeze event and on four occasions periodically after the freeze. A significant relationship (R2 = 0.33; p < .001) between freeze damage and MWT was observed. Prediction accuracies of freeze damage and MWT based on hyperspectral data varied among seedling portions (full-length, top, middle, and bottom portion of aboveground material) and scanning dates. Models based on the top portion were the most predictive of both freeze damage and MWT. The highest prediction accuracy of MWT [RPD (ratio of prediction to deviation) = 2.12, R2 = 0.78] was achieved using hyperspectral data obtained prior to the freeze event. Adoption of this assessment method would greatly facilitate the characterization and deployment of well-adapted loblolly pine families across the landscape.}, number={3}, journal={FOREST SCIENCE}, publisher={Oxford University Press (OUP)}, author={Lu, Yuzhen and Walker, Trevor D. and Acosta, Juan J. and Young, Sierra and Pandey, Piyush and Heine, Austin J. and Payn, Kitt G.}, year={2021}, month={Jun}, pages={321–334} }
 @misc{zhang_igathinathane_li_cen_lu_flores_2020, title={Technology progress in mechanical harvest of fresh market apples}, volume={175}, ISSN={["1872-7107"]}, DOI={10.1016/j.compag.2020.105606}, abstractNote={This article reviews the research and development progress of mechanical harvest technologies for fresh market apples over the past decades with a focus on the predominant technologies of shake-and-catch, robots, and harvest-assist platform methods. In addition, based on the review it points out the bottlenecks and future trends of these three technology categories. Major progress in the shake-and-catch method is related to theoretical studies on the effective removal of apples and catching mechanisms to minimize bruising. The unacceptable bruising conditions hinder the shake-and-catch method from commercial application. Two startups of apple harvesting robots are in the stage of commercializing their products based on vacuum and three-finger end-effectors, respectively. Economic benefits, as well as technology reliability and robustness of both robots, are pending for validation before they are on the market. In addition, a key obstacle faced by both robots before commercial use is to find a solution to pick apples grown in clusters. Harvest-assist platforms are gradually adopted by apple growers, but at a very low rate due to their doubts on economic benefits. Validation of harvest-assist platforms’ economic benefits and incorporation with more functions (e.g., sorting) would enhance their adoption. With the rapid development of sensing and automation technologies, such as novel sensors, embedded systems, and machine learning algorithms, and the progress in new tree canopy structures that are friendlier for fruit visibility and accessibility, it is believed the robots for fresh market apple harvest would be realized and commercialized in the near future. Currently, more efforts should be invested in analyzing and validating the economic benefits of harvest-assist platforms, as well as adding more functions to the harvest-assist platforms, to increase their application rate for the benefit of the apple industry.}, journal={COMPUTERS AND ELECTRONICS IN AGRICULTURE}, author={Zhang, Z. and Igathinathane, C. and Li, J. and Cen, H. and Lu, Y. and Flores, P.}, year={2020}, month={Aug} }