Project : USDA ARS

ARS Home » Plains Area » Lubbock, Texas » Cropping Systems Research Laboratory » Cotton Production and Processing Research » Research » Research Project #437975

Research Project: Enhancing the Profitability and Sustainability of Upland Cotton, Cottonseed, and Agricultural Byproducts through Improvements in Pre-Ginning, Ginning, and Post-Ginning Processes

Location: Cotton Production and Processing Research

2025 Annual Report

Objectives
OBJECTIVE 1: Develop commercially viable methods and technologies for use before ginning that reduce harvest costs, preserve fiber/seed quality, enhance the utilization of production/harvest/gin data, and prevent/minimize contamination of upland cotton. Subobjective 1A: Assessing the influence of seed cotton storage in round modules on lint and seed quality. Subobjective 1B: Improving the cleanliness and quality of stripper-harvested cotton through improved field cleaning systems. Subobjective 1C: Development of equipment to detect and remove contaminants from cotton during the harvesting process. OBJECTIVE 2: Enable commercially preferred technologies/methods/strategies for use in ginning upland cotton that improve cleanliness of seed cotton and lint, detect/remove contamination, preserve fiber quality, and reduce financial costs. Subobjective 2A: Development of equipment to detect and remove contaminants from cotton in the harvest and ginning processes. Subobjective 2B: Improving cotton fiber length distribution through novel lint cleaner design. OBJECTIVE 3: Develop commercially viable post-ginning technologies/techniques that enhance the storage and utilization of upland cotton products/coproducts/byproducts and reduce the environmental footprint of cotton production/processing. Subobjective 3A: Development of a commercially viable mechanical cottonseed delinting system to remove cotton linters and produce planting quality seed, without the use of chemicals. Subobjective 3B: Reducing particulate emissions from cotton ginning through improved pollution abatement device design using computational fluid dynamics (CFD) and laboratory testing. Subobjective 3C: Develop and evaluate the use of cotton plant constituents and other natural fibers in the manufacture of composite materials.

Approach
This five-year project plan addresses critical pre-ginning, ginning and post-ginning issues facing cotton producers and processors in the United States. Our plan of work is based on an interactive research approach which is focused on the development of processes and systems for preserving cotton quality during infield storage and ginning, removing foreign material and contaminants from seed cotton during harvesting and ginning, reducing particulate emissions from ag operations, reducing the environmental impact of acid cottonseed delinting, and increasing the value of cotton byproducts though composite materials. The research plan detailed herein addresses the development of new technologies, methods, and strategies for reducing the economic and environmental costs of cotton harvest, ginning, and post-gin processing of upland cotton and cotton by-products. Commercial viability of the research is a key component of any problem solution.

Progress Report
This is the final report for project 3096-21410-009-000D. Research will continue under project 3096-30600-001-00D. Objective 1: 5-yr Summary: Field scale experiments were conducted in Mississippi and Texas on cylindrical or “round” modules formed from spindle-picked and machine-stripped cotton to investigate the relationship between the change in fiber and seed quality parameters as a function of seed cotton moisture content at harvest and storage period duration. Analysis of seed cotton samples collected before and after storage in round modules indicated that significant economic losses due to degradations in fiber yellowness, reflectance, and foreign matter content are likely to be realized when oven-based reference method seed cotton moisture content at harvest exceeds 11%. This threshold for safe storage in round modules appears to be independent of storage period duration but additional work is underway to confirm this finding. The accuracy of non-reference method devices that measure seed cotton moisture content in round modules using electrical resistance, capacitance, and micro-wave permittivity were evaluated in the field. Moisture readings from all these devices tracked oven-based reference seed cotton moisture content linearly but the relationships varied by device. Thus, safe storage seed cotton moisture content thresholds varied by device. Additional analysis of seed data on oil quantity and quality and visible mechanical damage is underway. Scientific presentations to stakeholder groups have been given to report the results of this work. A new field cleaner was designed and implemented on commercial state-of-the-art cotton stripper harvesters. The new field cleaner exhibited 15% greater cleaning efficiency and 25% higher processing capacity than the field cleaner used on model year 2023 and older harvesters. Laboratory and field-based experiments on machine configuration and operation factors such as saw to grid bar clearances, grid bar spacing, and saw speeds were carried out to optimize cleaning efficiency and seed cotton loss. The results of these experiments have been documented in technical reports and presentations to the research partner and were implemented on new machines. A new active laydown system was designed and tested to help further improve the cleaning performance of the new field cleaner. The new system uses a powered cylinder with semi-rigid elements positioned around its periphery to increase the engagement of seed cotton by the saw teeth on the primary cleaning saw cylinder and actively bypass unengaged seed cotton to the middle saw cylinder. Testing revealed that the active laydown system allowed for precise control of the seed cotton loading rate on the top two saw cylinders in the field cleaner and helped to mitigate machine-plugging resulting from transient periods of overloading. The active laydown system improved cleaning efficiency by 5% compared to the production model field cleaner configuration but additional field testing is needed to confirm these results. Protocols for evaluating the presence of plastic contamination immediately in front of a harvester were successfully developed and validated. A mechanical exclusion system was designed and fabricated to operate in tandem with a smart machine-vision sensor, which was likewise designed and fabricated for seamless integration. The machine-vision software—central to the detection system—was completed, with traditional machine-learning classifiers evaluated and found capable of reliable detection under normal situations but is plagued by outlier corner cases that dictated a new approach. To address these short-comings, deep-learning approaches were assessed, and a Vision-Transformer–based model was selected for its lighting independent accuracy in outdoor conditions to handle these outlier corner cases. This model achieved performance metrics exceeding 95 % in independent testing and is now fully validated and ready for deployment and further research. Objective 2: 5-yr Summary: Plastic contamination detection and removal systems were designed and fabricated, and several commercial trials were conducted. Testing and evaluation of the system were completed in laboratory studies using a cut-down extractor feeder with commercial-scale cross-sectional geometry. Further studies in commercial cotton gins demonstrated that initial test results were successful for the primary sources of plastic contamination—specifically, the types that comprised more than 85% of contamination found at the USDA – Agricultural Marketing Service classing offices. The research revealed a significant impediment to technology adoption due to a lack of skilled personnel available to run the system. To address this issue, an auto-calibration system was developed and successfully completed with high-accuracy (> 95%) artificial intelligence (AI) models capable of detecting plastic contamination, as benchmarked against previously obtained commercial field data. Several high-speed AI models were developed to support this auto-calibration system, and scientific presentations reported on the new models and algorithms. In parallel, the authors developed a semi-automated workflow in which general-purpose Vision-Transformer-to-Caption models were paired with a lightweight semantic classifier to transform image captions into four target classes related to plastic contamination in cotton processing. This method was pursued to overcome the high cost and time required for manually labeling large datasets used to train specialized AI models in a domain where plastic contamination cost the industry an estimated $750 million annually. When evaluated on over 2,000 independent images, the combined system achieved more than 95% accuracy, enabling the offline annotation of over 50,000 images with minimal expert input. This novel approach saved over a year of man-hours and enabled rapid progress on the project. As a result, high-quality training sets were generated rapidly and used to train edge-deployable Vision-Transformer models for real-time plastic detection and cotton calibration. This approach reduced annotation and operational costs by orders of magnitude and provided a fully validated, scalable framework that can be adapted to other niche industrial applications. Utilizing this approach a massive image dataset was created and published and then used to train high-speed Vision-Transformer AI models that were then successfully deployed into several commercial trials in the 2024-2025 cotton ginning season. A novel multi-stage air-type lint cleaner was developed and tested for use in processing small lint samples from breeding and agronomic development research. The lint cleaner contains three cleaning points configured in a serpentine arrangement to remove foreign matter from lint. Foreign matter is removed through an opening in the flow duct via inertial force as the cotton lint (with entrained foreign material) passes at high speed around a sharp turn at each cleaning point. Elements of the design of the multi-stage air-type lint cleaner have been included in a U.S. patent and the performance of the machine was documented and the results communicated to the research partner through technical presentations and reports. Additional work to scale up the design for use under commercial ginning conditions is underway. Objective 3: 5-yr Summary: A commercial-scale mechanical delinter was built and installed in a commercial cotton gin processing both upland and extra long staple (ELS) cotton. The mechanical delinter was operated by the cotton gins personnel to delint organically grown cotton from some of their major growers. The system operated successfully and the seed was used by the producers to plant the following years crop. Both the drawings of the mechanical delinter and the results from the commercial trials have been documented and reported to stakeholders. Currently, there are stakeholders working to build similar units based on our design. Baffle-type pre-separators (BTPS) are used at some cotton gins to remove large trash particles from exhaust airflow streams and balance the air flow handled by commonly sized cyclones installed downstream. Both of these performance aspects of BTPS improve cyclone collection efficiency, thereby reducing particulate discharge into the environment. Previous two-dimensional computational fluid dynamics (CFD) modeling research on BTPS indicated that use of a secondary plate inside the settling chamber located just before the air discharge may improve particulate collection efficiency. To confirm this hypothesis, a new CFD model using Large Eddy Simulation to better characterize the interaction between air flow and particulate loading was developed. The new model revealed pulsed-flow characteristics not seen in previous models that help to explain some of the unexpected performance observed in laboratory testing. The results of this work have been communicated to stakeholder groups. Ten cellulose nanocrystal composites (CNC’s) were produced and tested at a collaborator’s facilities. The samples evaluated various combinations of ultrasonic treatments to improve thermal stability, mechanical properties, and fire resistance of the bio-based composites. Results showed the treatments with the addition of nano metallic fillers such as zinc oxide and boron oxide significantly improving fire resistance along with mechanical properties. Improvements in storage modulus associated with sonification amplitude and time were also noted up to 400 percent compared to the control.

Accomplishments
1. An automated system for generating training datasets for artificial intelligence. The single greatest barrier to developing and deploying custom artificial intelligence (AI) solutions is the training data. Creating the massive, labeled datasets required for training is a painfully slow and expensive manual process. For a typical industrial AI project, this means a human expert must spend over 520 workdays, more than two full years of effort, to manually annotate a 100,000 image dataset. This data bottleneck makes rapid AI development virtually impossible. ARS scientists in Lubbock, Texas, built the solution to this fundamental problem: a platform that automates the creation of AI training data. After an expert provides a small initial sample for guidance, the system becomes a fully automated engine, generating vast and accurately labeled datasets on its own. It eliminates the hundreds of days of manual labor that hold AI projects back. To validate our platform, we targeted a complex industrial challenge: developing an AI to detect plastic contamination in cotton, a problem that is estimated to cost the U.S. cotton industry approximately $750 million annually. Our system successfully generated the required 100,000-image dataset with over 95% accuracy, slashing the development timeline from years to a matter of days. The key accomplishment is not just the final plastic-detection AI, but the underlying technology that makes building such solutions fast, affordable, and scalable for the first time. Manually creating training data for an AI is like building a car engine by hand—a task that can take an excessive amount of time. We didn't just build a better data annotation engine; we invented the automated assembly line for it.

2. A system for managing cotton modules and harvest data using radio frequency identification. New cotton harvesters that form cylindrical or “round” modules cover the cotton in a multi-layered plastic film to contain and protect the cotton from rain, wind, and other environmental hazards encountered during storage. Imbedded in each module wrap are four radio frequency identification (RFID) tags that return a 24 character string that uniquely identifies each round module of cotton. Harvest data, specific to each round module collected on the harvester, are linked to each module in a database using the module identifier found on the RFID tags. ARS engineers at Lubbock, Texas, developed the Electronic Module Management (EMM) system that contains a set of hardware and software tools that allow cotton growers and ginners to automatically retrieve module specific harvest data from harvesters, track the physical position and processing status of each module as it moves from the harvester through the ginning process using RFID tags, and associate lint bales to the round modules from which they were ginned to facilitate site-specific mapping of both yield and quality. Software developed for the EMM system was developed under an open-source license and can be freely distributed to cotton growers and ginners for use “as-is” or used in a software development project custom tailored to the needs of their operation. Use of the EMM system reduces labor costs associated with managing modules at the gin and provides new critical data to growers looking to implement growing practices that manage yield and fiber quality on a site specific basis.

U.S. DEPARTMENT OF AGRICULTURE

Cotton Production and Processing Research: Lubbock, TX