Progress 07/01/23 to 02/29/24

Outputs
Target Audience:The primary target audience for this project is peach farmers in South Carolina, Georgia, Alabama, Arkansas, Tennessee, North Carolina, Florida, and California. In addition to peaches, other specialty fruit crops that require thinning are nectarines, apricots, plums, and apples, and farmers of these crops will also be a target audience. In addition, the vision-assisted AI can be trained to thin both the blooms and the fruits for these crops, as well as vegetables like carrots, beets, turnips, radishes, spinach, cilantro, parsnips, etc.The A-EYEtechnology can also be adapted for early detection of disease in fruit trees, orchards, and other specialty crops. In addition to the agriculture industry,the A-EYE technology can also be adapted and integrated with existing vision sensors in manufacturing and surveillance industries, including autonomous vehicles for identifying road activities and unmanned aerial vehicles for search-and-rescue. These industries would be future target audiences. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?University of South Carolina students were trained to analyze images of peach trees and identify blooms along the branches. The students used their training to annotate hundreds of images and place bounding boxes around thousands of peach blooms to support the computer vision models. Advent engineers were trained by farmers in the art and science of peach bloom thinning, learning the steps and thought processes that experienced farmers use to make thinning decisions. This knowledge was then digitized and used to develop a rules-based expert system to provide real-time thinning recommendations. How have the results been disseminated to communities of interest?A summary of this Phase I project and outcomes of the research and development have been disseminated to several recipients. In addition to the University of South Carolina and Titan Farms (who are both project team members), the research and development work has also been shared with the South Carolina Research Authority (SCRA), the South Carolina Department of Agriculture (SCDA), the SCDA's Agribusiness Center for Research & Entrepreneurship (ACRE), Palmetto Agribusiness Council, Clemson University Center for Agriculture Technology (CU-CAT), and South Carolina and Georgia farms, including Big Smile Peaches, Black's Peaches, Shuler Peaches, Heritage Farms, McLeod Farms, Jerrold A. Watson and Sons, Chappell Farms, Dickey Farms, Pearson Farms, and WP Rawl. Advent has also reached out to farms and organizations in California, including Masumoto Farms, Brandt Farms, Dry Creek Peach Farms, Hudson Farms, and the Almond Board of California. We presented the work to these groups to extend our contacts in the agriculture industry, get feedback on the technology, and explore future collaborations. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? 1. Develop computer vision system for estimating peach bloom density and distribution in real-time: The approach to meet this objective was to utilize a color calibration chart (SpyderCHECKR 24) to help calibrate to a consistent color appearance under different lighting conditions. Prior to processing, the drone/phone/tablet camera captures an image of a SpyderCHECKR 24 chart, and a convolutional neural network is used to detect the chart and identify its orientation. A color calibration matrix (CCM) is then automatically derived from the chart's color patches and then applied to every pixel in all subsequent images/video frames captured during that session. The color calibration and correction technique enables accurate measurement of peach bloom density and distribution on individual trees, lots, and entire orchards from images and streaming video. 2. Build an intelligent software to classify the peach bloom that requires thinning: This objective was achieved by employing an object detection procedure using a deep convolutional neural network to detect peach blooms. Once detected, an algorithm was developed to determine the depth and 3D spatial location of each bloom. The 3D spatial location of each bloom is used to distinguish blooms belonging to a tree of interest from those on trees in the background. This methodology enables the ability to obtain rapid bloom counts on individual branches, trees, rows, and lots, which farm managers need to prioritize thinning and maintenance activities and estimate crop yield. 3. Provide recommendations for thinning action -- Go (keep the bloom) and No-go (remove the bloom): This objective was achieved by using the bloom density, distribution, count, and spatial location from above, along with a human-based decision-making process for thinning by the farmers. Using machine learning, an AI system was developed based on a rule-based expert system to provide autonomous thinning recommendations in real-time. The recommendations are displayed in real-time in an augmented reality interface that provides the total bloom count and highlights which blooms to remove to maximize production. The A-EYE software is expected to eliminate the training time and considerably reduce the thinning time, reducing the expenses of peach farms by 50% while bringing a new capability to the market.

Publications

• Type: Other Status: Accepted Year Published: 2024 Citation: Beard et al., "Computer Vision and Artificial Intelligence for Crop Thinning," Provisional Patent Application, No. 63/664,214, Jun 2024.