Water used for blanching, post-blanching cooling, and cooking of fruits and vegetables can, in general, be collected and reused for the initial washing of incoming products without treatment . Reuse of flume water. Instead of discharging flume water to the wastewater stream, it can be recovered, filtered, and reused continuously in fluming applications. Alternatively, flume water can be recovered and recycled for use in equipment pre-rinsing and pre-soaking applications elsewhere in the facility . Reuse of compressor cooling water. Cooling water from compressors can be reused as seal water in vacuum pumps instead of fresh water, or as secondary water for other purposes, such as equipment pre-soaking . Warm cooling water can also be stored in insulated tanks for later use in facility cleaning, pre-soaking, and equipment pre-rinsing applications .The U.S. fruit and vegetable processing industry spent nearly $810 million on purchased fuels and electricity in 2002, making energy a significant cost driver for the industry. Energy efficiency improvement is an important way to reduce these costs and to increase predictable earnings in the face of ongoing energy price volatility. Considering the negative impacts of the 2001 spike in U.S. natural gas prices on the industry’s operating costs, stackable planters as well as more recent sharp increases in natural gas prices across the nation, energy efficiency improvements are needed today more than ever.
Many companies in the U.S. fruit and vegetable processing industry have already accepted the challenge to improve their energy efficiency in the face of high energy costs and have begun to reap the rewards of energy efficiency investments. This Energy Guide has summarized a large number of energy-efficient technologies and practices that are proven, cost-effective, and available for implementation today. Energy efficiency improvement opportunities have been discussed that are applicable at the component, process, facility, and organizational levels. Preliminary estimates of savings in energy and energy-related costs have been provided for many energy efficiency measures, based on case study data from real-world industrial applications. Additionally, typical investment payback periods and references to further information in the technical literature have been provided, when available. A key first step in any energy improvement initiative is to establish a focused and strategic energy management program, as depicted in Figure 6.1, which will help to identify and implement energy efficiency measures and practices across and organization and ensure continuous improvement. Tables 5.1 to 5.3 summarize the energy efficiency measures presented in this Energy Guide. While the expected savings associated with some of the individual measures in Tables 5.1 to 5.3 may be relatively small, the cumulative effect of these measures across an entire plant may potentially be quite large. Many of the measures in Tables 5.1 to 5.3 have relatively short payback periods and are therefore attractive economic investments on their own merit.
The degree of implementation of these measures will vary by plant and end use; continuous evaluation of these measures will help to identify further cost savings in ongoing energy management programs. In recognition of the importance of water as a resource in the U.S. fruit and vegetable processing industry, as well as its rising costs, this Energy Guide has also provided information on basic, proven measures for improving plant-level water efficiency, which are summarized in Table 5.4. For all energy and water efficiency measures presented in this Energy Guide, individual plants should pursue further research on the economics of the measures, as well as on the applicability of different measures to their own unique production practices, in order to assess the feasibility of measure implementation.This work was supported by the Climate Protection Partnerships Division of the U.S. Environmental Protection Agency as part of its ENERGY STAR program through the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. Many people within and outside of the fruit and vegetable processing industry provided valuable insights in the preparation of this Energy Guide. The authors are particularly grateful to the following people for their helpful comments and advice during the development of this Energy Guide: John Batura of Campbell’s Soup Company, Joseph Benevides of Ocean Spray Cranberries, Austin H. Bonnett , Alan Christie of the J.R. Simplot Company, Elizabeth Dutrow of the U.S. Environmental Protection Agency, Dan Fonner of Heinz North America, Paul Halberstadt of ConAgra Foods, Al Halvorsen of Frito-Lay, Don Hertkorn of ICF Consulting, Ted Jones and Ilene Mason of the Consortium for Energy Efficiency, John Malinowski of Baldor Electric Company, Aimee McKane of Lawrence Berkeley National Laboratory, Leland McPherrin of Escalon Premier Brands, Rob Neenan of the California League of Food Processors, Richard Pate of the Pacific Gas and Electric Company, Linda Raynes of the Electrical Apparatus Service Association, Jon Russett of General Mills, and Terry Young of ConAgra Foods.
Any remaining errors in this Energy Guide are the responsibility of the authors. The views expressed in this Energy Guide do not necessarily reflect those of the U.S. Environmental Protection Agency, the U.S. Department of Energy, or the U.S. Government. We discovered that there are only a few categories of interest in a highly domesticated breeding population and that a small number of features are that ordinal shape categories are highly heritable and that the features needed for accurate classification are also heritable.The data released with this article contain digital images of 6,874 strawberry fruit from 572 hybrids originating from the University of California, Davis, Strawberry Breeding Program. The data for this article, including pre-processed images , processed images , and extracted features , are available on Zenodo. The pre-processed images typically contained multiple berries per image along with a data matrix bar code indicating the genotype ID and other elements of the experiment design. The processed images are 1,000 × 1,000 pixels-scaled binary images of individual fruit. The extracted features data set is provided as a CSV file. Additionally, snapshots of the code and data supporting this work are available in the GigaScience repository, GigaDB. We hope that the release of these data assists others in developing novel morphometric approaches to better understand the genetic, developmental, and environmental control of fruit shape in strawberry, and more broadly in other fruits, vegetables, and specialty crops.k-Means clustering rapidly detects patterns in large, multidimensional data sets used for clustering, decision making, and dimension reduction. It is an iterative algorithm that partitions a data set into a pre-defined number of non-overlapping clusters, k, by minimizing the sum of squared distances from each data point to the cluster centroid. A centroid corresponds to the mean of all points assigned to the cluster. Here, we used k-means to cluster flattened binary images . Individual fruits were segmented from the image background as a binary mask, normalized by the major axis, resized to 100 × 100 pixels, and flattened into a vector . We represented each image as a 10,000-element vector containing binary pixel values. We were able to rapidly and reliably assign images to classes using k-means clustering. In this experiment, we allowed k, the number of permitted categories, to range from 2 to 10. This range was chosen because we anticipate that a human-based classification system would not have the speed or reliability needed for this task, particularly for larger values of k.As high-throughput phenotyping for external fruit characteristics becomes of interest to specialty crop researchers, we expect that this work will have various applications in both applied and basic plant research, stacking pots intellectual property protection and documentation, and waste reduction. Our study showed that strawberry fruit shapes could be robustly quantified and accurately classified from digital images. Most importantly, our analyses yielded quantitative phenotypic variables that describe fruit shape , arise from continuous distributions, and are moderately to highly heritable . We accomplished this by translating 2D, digital images of fruit into categorical and continuous phenotypic variables using unsupervised machine learning and morphometrics. We found that mathematical approaches developed for human face recognition were powerful for strawberry fruit shape phenotyping , that unsupervised shape clustering was robust to sample size deviations, and that only a few quantitative features are needed to accurately classify shapes from images , indicating a paradigm appropriate for genetic dissection.
Digital plant phenotyping is able to empower quantitative genetic analyses by providing heritable and biologically relevant, latent phenotypes in a cost-effective manner. In many cases, these latent traits are derived from PCA, multidimensional scaling , structured equation modeling , persistent homology , or auto-encoding convolutional neural networks, which can be exceedingly abstract and difficult to interpret biologically but may also reveal unexpected patterns of phenotypic and genetic variation. Many of the features described in this study, along with those reported by Turner et al. and Gage et al. , had high heritability and are exciting targets for future quantitative genetic analyses, including GWAS and genomic prediction, which have been shown to be successful for shape features in recent work in rice, apple, and pear. However, the H2 of 1 selected feature in this study, EigenFruitPC3, was estimated to be 0.00. Similar results were reported in carrot for pixel-based root and shoot features, apple for elliptical Fourier leaf shape features , and corn for pixel-based shoot features. Turner et al. attributed the null H2 of root shape characteristics to low phenotypic variation between the inbred parents and genotype × environment interactions. This pattern, while seemingly present, was not discussed in detail by either Migicovsy et al. or Gage et al.. While there may be many drivers for this pattern, we hypothesize that the null estimate may arise from the pixel-based descriptors describing more complex aspects of fruit or root shape. If the non-genetic component of a multivariate phenotype is large, then performing PCA on that multivariate trait could produce leading PCs that describe mostly non-genetic variance . However, there are too few reports to adequately determine the likelihood and causal source of this phenomenon. We empirically derived the shape progression produced in the present study through the application of a new method, PPKC, and used these mathematical categories to interpret the extracted shape features . Ordinal categorical traits are commonplace in quantitative genetic studies, a current standard for phenotyping external fruit characteristics, and enable understanding and explanation of complex, latent space plant phenotypes . PPKC specifically considers the relationship between a cluster at k and all clusters for values <k as a covariance matrix and projects this k-dimensional space to 1 dimension using eigen decomposition. Ordination using dimension reduction techniques, including PCA, correspondence analysis, and MDS has been previously proposed and used in community ecology. Theoretically, the eigen decomposition step of PPKC could be replaced with another technique. However, unlike methods using eigen decomposition, which progressively subdivides variation such that the position on the leading axis is fixed regard-less of the number of axes examined, the position of samples on MDS axes may change when different dimensions are extracted, making MDS axes arbitrary and without meaning other than a convenient reference. PPKC identified 4 exemplary strawberry shape categories in the population that we studied, which were characterized by a progression from ”longer-than wide” to ”wider-than-long”. This ordinal scale can be used in breeding and research programs as traits of interest, or it can be used to organize and interpret more abstract quantitative features, such as EigenFruitPCs or SEM latent variables, through supervised machine learning algorithms. Critically, this gradient agreed with the arbitrarily defined progressions in previous reports. However, unlike previous studies, which suggested using 9 ordinal or11 nominal shape categories, our work presented empirical evidence for a smaller number of mathematically defined shape categories. We determined that k = 4 was the appropriate level of complexity on the basis of the visual appearance of the discovered clusters , high H2 estimates , and the information criteria calculated for the k-means models. Interestingly, PPKC can determine a visually, reasonable phenotypic gradient up to k = 8 despite strong evidence of overfitting for k > 4. We extrapolate that PPKC should continue to work beyond k = 9 so long as new clusters are distinct and do not arise as an artifact of overfitting k. The specific genetic factors that give rise to variation in fruit shape in octoploid, garden strawberry are currently unclear or understudied. The selective pressure exerted on fruit shape in strawberry could have affected large-effect loci, in which case ordinal phenotypic scores are likely to be sufficient for identifying genetic factors affecting fruit shape. Loss- and gain-of function mutations have played an essential role in identifying genes affecting fruit shape in tomato, a model that has been highly instructive and important for understanding the genetics of fruit shape and enlargement in plants.