SBIR-STTR Award

Large commercial breeding layer broiler turkey and production flocks collect an overwhelming amount of data much of which is either inaccurate and/or not leveraged for decision making. Current poultry data management solutions are typically ineffective poorly integrated across the supply chain (i.e. breeders to processing plants) and ultimately force farmers to make production food safety and economic decisions based on incomplete and inaccurate data. Hence the ability to accurately collect and analyze data in order to be predictive is seen as the "holy grail" of poultry production efficiency poultry health and food safety. From an innovation perspective this will require improved and greater utilization of remote sensing in addition to optimization of various predictive models. As some companies start to explore Machine Learning (ML) based approaches(i.e. statistical techniques that are capable of "learning by finding" non-obvious associations and patterns in the data in order to create more reliable custom accurate explanatory and predictive statistical models) for predictive analyses there is a general lack of knowledge about which ML approaches 'work' and how ML based results can be integrated within decision science based tools which highlight company expertise as opposed to data. In other words the ability to create integrative software that stresses both novel predictive statistics and institutional knowledge would allow data and expert knowledge to be dovetailed and considered equally by decision makers. One hindrance to this approach is the inability to acquire accurate data. "Garbage in garbage out" is a real problem in the analysis of ag-based data including poultry data from the farm to the processing plant. One paramount example is the inability to count eggs at the poultry house level(as opposed to the processing plant). Without these data at the house level counting is only done accurately at the processing plant which makes flock management at the house and row level impossible. Here we propose to develop two innovations and dovetail the technologies.1. With respect to egg production we propose to further develop our novel modular egg counter that attaches to commercial egg conveyor belts of differing width and detects egg counts using different approaches including ultrasonic analysis infrared analysis and image analysis.2. With respect to predictive data analysis and decision sciences for layer broiler and turkey production using ML based techniques we propose to test 3 ML based predictive approaches from historic raw data provided by various commercial poultry partners we are affiliated with including our first commercial client Hy-Line North America (an international layer genetic company). Our goal is to identify the best predictive ML based model(s) to better predict production economics and food safety outcomes. The analyses from the ML will be further leveraged by the incorporation of decision science based tools like Analytical Hierarchy Processes (AHP) that can help poultry companies quantitatively query internal experts in order to integrate expert opinion within a company to statistical observation and cost-benefit analysis provided by ML analysis based tools. The greatest improvements in poultry production efficiency food safety and economics will be made via the improved accuracy of data and integration and utilization of data. As experts in poultry data analytics at the academic and industry level Agrinerds founders recognize the problems and potential innovations described above. Our goal for this phase I Program is to develop a commercially viable innovative methodology for egg collection (hardware) and link that to innovative methods of analysis (software) to better predict outcomes for the commercial poultry industry. Our commercial client Hy-Line and collaborators Purdue Chicken and JSWest are major producers that giv