1<html class="" lang="en" dir="ltr" data-whatinput="mouse" data-whatintent="mouse"><head> 2 <meta charset="utf-8"> 3 <meta http-equiv="x-ua-compatible" content="ie=edge"> 4 <meta name="viewport" content="width=device-width, initial-scale=1.0"> 5 <title>AI on the Bog</title> 6 <link rel="stylesheet" href="https://d33wubrfki0l68.cloudfront.net/bundles/52488c77a948cb7a211664741de9c7ac101496c3.css"> 7 8 <link rel="icon" type="image/svg+xml" href="./../../icons/cranberry.svg" /> 9 <link rel="âpreconnectâ" href="âhttps://www.google-analytics.comâ">
vendor: 67 bytes, lines 9-10
9 10<script async="" src="https://www.googletagmanager.com/gtag/js?id=
10UA-146060534-1
vendor: 12 bytes, line 10
10"></script>
11<script> 12
vendor: 133 bytes, lines 12-15
12window.dataLayer = window.dataLayer || []; 13 function gtag(){dataLayer.push(arguments);} 14 gtag('js', new Date()); 15 gtag('config', '
15UA-146060534-1
vendor: 4 bytes, line 15
15');
16</script>
16<meta class="foundation-mq"></head> 17 <body> 18 19 <header data-sticky-container="" class="sticky-container" style="height: 42px;"> 20 <div class="sticky sticky-topbar is-anchored is-at-top" data-sticky="5hmg58-sticky" data-options="anchor: page; marginTop: 0; stickyOn: small;" data-resize="x52oh8-sticky" data-mutate="x52oh8-sticky" data-events="resize" style="max-width: 346px; margin-top: 0px; bottom: auto; top: 0px;"> 21 <div class="title-bar" data-responsive-toggle="top-bar" data-hide-for="medium" style=""> 22 <button class="menu-icon" type="button" aria-label="Menu" data-toggle=""></button> 23 <div class="title-bar-title">menu</div> 24 </div> 25 <div id="top-bar" class="top-bar" style="display: none;"> 26 <ul class="menu vertical medium-horizontal dropdown" data-responsive-menu="medium-dropdown" data-options="animationEasing:swing; animationDuration:750" role="menubar" data-dropdown-menu="ihvoaq-dropdown-menu" data-mutate="sv07ce-responsive-menu"> 27 <li role="menuitem"><a href="../../index">research</a></li> 28 <li role="menuitem"><a href="../../photo/index">travel+photography</a></li> 29 <li role="menuitem"><a href="../../about/index">about</a></li> 30 </ul> 31 </div> 32 </div> 33 </header> 34 35 <article class="grid-container"> 36 <div class="grid-x grid-margin-x align-justify"> 37 <div class="large-9 medium-9 cell"> 38 <h1>AI on the Bog: Monitoring and Evaluating Cranberry Crop Risk</span></h1> 39 <p class="subtitle"><i>Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2021</i></p> 40 <h5><a href="https://periakiva.github.io/" target="_blank"> Peri Akiva</a>, Benjamin Planche, Aditi Roy, <a href="https://www.ece.rutgers.edu/~kdana/" target="_blank"> Kristin Dana</a>, <a href="https://pemaruccicenter.rutgers.edu/people/oudemans.html" target="_blank"> Peter Oudemans, Michael Mars</a></h5> 41 42 <hr> 43 <span class="tag-wrapper"><span id="abstract" data-magellan-target="abstract" class="tag"></span></span> 44 <article> 45 <center> 46 <div class="media-object"> 47 <div class="media-object-section"> 48 <td width=150px> 49 <center> 50 <img style="width:800px" srcset="./resources/WACV2021_overall_pipeline_round_corners-1.png"> 51 <p class="image-caption">Pipeline overview. Cloud image sequence, humidity, and wind speed are used to predict 52 future berry temperature over a time horizon to determine high risk time periods. Aerial imagery of cranberry 53 crops are used to obtain count density maps to determine high risk regions. Exposure metrics (number of exposed 54 cranberries with high berry internal temperature) are made available to the farmer in order to make resource 55 decision such as crop irrigation. Red dashed boxes indicate high risk regions. Best viewed in color and zoomed.</p> 56 </center> 57 </td> 58 </div> 59 </div> 60 <div class="media-object"> 61 <div class="media-object-section"> 62 <td width=200px> 63 <center> 64 <img style="width:800px" srcset="./resources/figure_pipeline_cloud_seg_and_motion_est-1.png"> 65 <p class="image-caption">Training of baseline method for joint semantic segmentation and optical flow regression with additional self-supervision.</p> 66 </center> 67 </td> 68 </div> 69 </div> 70 <div class="media-object"> 71 <div class="media-object-section"> 72 <td width=200px> 73 <center> 74 <img style="width:800px" srcset="./resources/WACV2021_cranberry_seg_count_architecture-1.png"> 75 <p class="image-caption">Triple-S Network architecture used in this application. 76 Aerial image is input to a U-Net style network with output guided bysegmentation loss, 77 L<sub>seg</sub>, split loss, L<sub>split</sub>, and count loss, L<sub>count</sub>. 78 W<sub>select</sub> is the selective watershed algorithm 79 introduced in [1], and CC is the connected components algorithm.</p> 80 </center> 81 </td> 82 </div> 83 </div> 84 <div class="media-object"> 85 <div class="media-object-section">
86 <td width=200px> 87 <center> 88 <img style="width:650px" srcset="./resources/segmentation_SCCP.png"> 89 <p class="image-caption">Cloud future segmentation results.</p> 90 </center> 91 </td> 92 </div> 93 </div> 94 </center> 95 <h3>Abstract</h3> 96 <p>Machine vision for precision agriculture has attracted considerable research interest in recent years. 97 The goal of this paper is to develop an end-end cranberry health monitoring system to enable and support 98 real time cranberry over-heating assessment to facilitate informed decisions that may sustain the economic 99 viability of the farm. Toward this goal, we propose two main deep learning-based modules for: 1) cranberry 100 fruit segmentation to delineate the exact fruit regions in the cranberry field image that are exposed to sun, 101 2) prediction of cloud coverage conditions to estimate the inner temperature of exposed cranberries. 102 We develop drone-based field data and ground-based sky data collection systems to collect video imagery at 103 multiple time points for use in crop health analysis. Extensive evaluation on the data set shows that it is 104 possible to predict exposed fruitâs inner temperature with high accuracy (0.02% MAPE) when irradiance is 105 predicted with 8.41-20.36% MAPE in the 5-20 minutes time horizon. With 62.54% mIoU for segmentation and 106 13.46 MAE for counting accuracies in exposed fruit identification, this system is capable of giving informed 107 feedback to growers to take precautionary action (\eg, irrigation) in identified crop field regions with higher 108 risk of sunburn in the near future. Though this novel system is applied for cranberry health monitoring, it 109 represents a pioneering step forward in efficiency for farming and is useful in precision agriculture beyond 110 the problem of cranberry overheating.</p> 111 </article> 112 113 <hr> 114 <span class="tag-wrapper"><span id="video" data-magellan-target="video" class="tag"></span></span> 115 <article> 116 <h3>Video</h3>
117 <div class="flex-video widescreen"> 118 <iframe width="560" height="315" src="https://www.youtube.com/embed/lp3puaLfvUs" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> 119 </div> 120 </article> 121 <hr> 122 <hr> 123 <span class="tag-wrapper"><span id="paper" data-magellan-target="paper" class="tag"></span></span> 124 <div class="bottom-section"> 125 <article> 126 <h3>Paper</h3> 127 <div class="media-object stack-for-small"> 128 <div class="media-object-section"> 129 <a href="https://arxiv.org/pdf/2011.04064.pdf" target="_blank"> 130 <img class="thumbnail" style="width:350px" srcset="./resources/paperfront.png"> 131 </a> 132 <p class="image-caption">Paper and Supplement (Arxiv)</p> 133 </div> 134 </div> 135 <p data-tooltip="kgaqi1-tooltip" class="bibtex btn thumbnail has-tip" title="" data-clipboard-text="@misc{akiva2020h2onet, 136 title={H2O-Net: Self-Supervised Flood Segmentation via Adversarial Domain Adaptation and Label Refinement}, 137 author={Peri Akiva and Matthew Purri and Kristin Dana and Beth Tellman and Tyler Anderson}, 138 year={2020}, 139 eprint={2010.05309}, 140 archivePrefix={arXiv}, 141 primaryClass={cs.CV} 142 }" aria-describedby="ue88yi-tooltip" data-yeti-box="ue88yi-tooltip" data-toggle="ue88yi-tooltip" data-resize="ue88yi-tooltip" data-events="resize"> 143 @misc{akiva2020ai, <br> 144 title={AI on the Bog: Monitoring and Evaluating Cranberry Crop Risk}, <br> 145 author={Peri Akiva and Benjamin Planche and Aditi Roy and Kristin Dana and Peter Oudemans and Michael Mars}, <br> 146 eprint={2011.04064}, <br> 147 archivePrefix={arXiv}, <br> 148 primaryClass={cs.CV} <br> 149 year = {2020} <br> 150 </p> 151 <span class="tag-wrapper"><span id="code" data-magellan-target="code" class="tag"></span></span> 152 <p> 153 <img src="https://d33wubrfki0l68.cloudfront.net/fabe53eb72f9b6d3d47cd95aff31ffc45c2fdbf8/5cde4/icon/github.png" height="25" width="25"> 154 <a href="link-to-github" target="_blank" style="display: inline-block; vertical-align: middle;"> 155 Open Source Code 156 </a> 157 158 </p> 159 </article> 160 </div> 161 </div> 162 163 <div id="sidebar" class="large-3 medium-3 cell show-for-medium sticky-container" data-sticky-container="" style="height: 220px;"> 164 <div id="sidebar-sticky" class="sticky is-anchored is-at-top" data-sticky="e5yfqz-sticky" data-margin-top="5.65" data-resize="sidebar-sticky" data-mutate="sidebar-sticky" style="max-width: 0px; margin-top: 0px; top: 0px; bottom: auto;" data-events="mutate"> 165 <h5>Navigation:</h5> 166 <ul class="dash nav medium-shrink-text" data-magellan="b1jzdo-magellan" data-resize="0dk5a5-magellan" data-scroll="0dk5a5-magellan" id="0dk5a5-magellan" data-events="resize"> 167 <li><a href="#abstract" class="">Abstract</a></li> 168 <li><a href="#video" class="">Video</a></li> 169 <li><a href="#paper" class="">Paper</a></li> 170 <li><a href="#code" class="">Code</a></li> 171 </ul> 172 </div> 173 </div> 174 175 </div> 176 </article> 177
177<script src="https://d33wubrfki0l68.cloudfront.net/bundles/291f53dcefa5226447d7ae63702f89c5390e2e71.js"></script>
177<div class="tooltip" role="tooltip" aria-hidden="true" data-is-active="false" data-is-focus="false" id="ue88yi-tooltip" style="display: none;">Copied!</div> 178
178<script>var clipboard = new ClipboardJS('.btn');</script>
178 179 180 181</body></html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.