Recipient Organization
NORTH DAKOTA STATE UNIV
1310 BOLLEY DR
FARGO,ND 58105-5750
Performing Department
Agricultural and Biosystems Engineering
Non Technical Summary
Digital agriculture is using advanced technologies to help farmers and stakeholders improve food production. Farmers are the "CEO" of their farm business. They need to make complex quick decisions based on experience, measurement and recommendations. With digital agriculture's help, farmers can gather data frequently and accurately to assist in making decisions. Technologies like artificial intelligence, robotics, sensors, and unmanned aerial systems (UAS) will bring valuable data collection/processing/visualization methods to help make more accurate farming decisions. Digital agriculture has the potential to make agriculture more productive, more consistent and more efficient. In the state of North Dakota, agriculture is a vital economic force. About 90 percent of the state's land is used for agriculture, totaling 40 million acres devoted to farming and ranching. Additionally, 24 percent of the state's population is employed in the agricultural sector alone. Plus, there are 1.7 million beef cattle, or almost three cattle for every North Dakotan. Therefore, implementing digital agriculture tools in the state of North Dakota will have numerous opportunities to benefit crop and livestock production. For crop production, weed control is a critical issue that has been exacerbated by the widespread occurrence of herbicide-resistant weeds. The cost of not using herbicide in US crop production would cause an income loss of $21 billion (Gianessi & Reigner, 2007). The incidence of herbicide resistance in the ND-MN region has increased from one case in 2006 to over 100 cases in 2013. Kochia, common ragweed, giant ragweed, waterhemp and horseweed are some of the most commonly sighted herbicide-resistant weeds. Most often, it is too late or too difficult to control the weeds by the time the resistant weeds are detected. If the herbicide-resistant weeds can be identified and mapped in a timely critical manner, farm managers could implement alternate weed management practices suitable for controlling herbicide-resistant weeds. Similarly, certain new weeds such as Palmer amaranth, have already appeared in North Dakota, which could render crop fields unsuitable for planting. Palmer amaranth is a cause of significant concern as it is difficult to control. For livestock production, beef cattle production is one of the top five economic drivers (wheat, cattle and calves, soybeans, corn for grain, and sugar beets) for North Dakota agriculture (NETSTATE, 2016). The ability to consistently deliver high-quality beef products to consumers must remain top-of-mind for North Dakota beef producers. "Taste" remains the primary decision factor driving beef purchase. Experimental market research has shown that beef consumers equate eating quality (tenderness, flavor, juiciness) with value and that superior eating quality not only increases the likelihood that consumers will purchase beef, but also results in an increased willingness to pay to obtain the level of eating satisfaction they desire (Tatum, et, al., 2017). Pre-harvest factors that affect beef quality attributes include those that are inherent to the animals themselves and those that are associated with cattle production practices (e.g., diet, time-on-feed, use of growth enhancement technologies, handling/stress, health). Beef cattle temperament (defined as the reactivity of cattle to humans) affects many physiological systems, including immunity, stress, and metabolism (Sanchez et al., 2016) as well as carcass characteristics and eating quality (Cafe et al., 2010; Francisco et al., 2015; Magolski et al., 2013). For example, Japanese Kobe beef "are treated like kings", where classical music is played at the production site as a relaxation technique at feeding time. The cattle associate the music with eating and hence the music improves their appetite (Moorhead, 2010; Korteman, 2013). Therefore, understanding and having the means to measure production factors that impact exceptional beef quality could influence beef production practices in North Dakota and open market opportunities to consumers seeking superior flavor and taste attributes from their beef products. Computer vision technologies have been developed and researched over the past few decades. It has become a popular tool in managing precision agriculture including the crop and livestock industry (Arias, Molina, & Gualdron, 2004; Patrício & Rieder, 2018; Rius-Vilarrasa, Bünger, Maltin, Matthews, & Roehe, 2009). In this research, we will develop different imaging technologies that are rapid, accurate, and targeted to identify weed species and beef cattle temperament to help North Dakota crop and livestock production.
Animal Health Component
40%
Research Effort Categories
Basic
20%
Applied
40%
Developmental
40%
Goals / Objectives
[Crop] Integration of thermal, visible/near-infrared (NIR) and hyperspectral sensing and analysis methods for weed species and herbicide resistance identification and mapping.[Livestock] Study of the relationship between beef cattle temperament and beef quality using video and thermal imaging technologies.
Project Methods
Identify weed species using thermal, visible and NIR sensing and analysis methodsA method will be developed based on greenhouse data and staged field plot data (by transplanting three species of weeds of same age into corn fields) to detect three common weed species and their herbicide resistance using thermal and visible/NIR sensor data. The thermal and visible/NIR data will be collected independently and models will be developed independently for species and resistance identification. We propose to mount thermal and visible/NIR cameras simultaneously on a UAS platform to collect simultaneous images of the field with 70-80% overlap and sub-centimeter spatial resolution. These images will be co-registered into geographical coordinates although the thermal camera has a much lower resolution than the visible/NIR cameras. A fluorescence camera will be used to collect data on chlorophyll fluorescence of resistant and susceptible plants at greenhouse and field conditions to evaluate whether this added characteristic will enhance classification accuracy, if included as an additional feature. The models for weed classification and herbicide resistance identification will be implemented on a co-registered image by following the ensuing steps:(a) Perform 3D structural analysis with a single visible/NIR band to develop a 3D point cloud.(b) Identify all plants as objects in the scene removing the remaining pixels as background.(c) Extract structure information such as canopy height to width ratio, branch angle and canopy density for all plant objects.(d) Perform an image segmentation based on the row location and plant structural variables to remove crop plants from the scene.(e) Combine structural variables with color values from the visible/NIR image or vegetation indices derived from the band numbers, and canopy temperature from thermal band for all weed objects.(f) Perform classification of weed plants using the combined color and structural variables to identify weed species, and thermal data to identify herbicide resistance.(g) Develop a map indicating weed species and their distribution, along with herbicide resistance status.Identify herbicide-resistant Palmer amaranth using visible/NIR, thermal and hyperspectral image analysis methods.In addition to the first procedure, we will collect two sets of data from one to two crop fields infested with Palmer Amaranth, one set before herbicide application and a second set after herbicide application. Field data collected will include spectral data, visible/NIR and thermal image data from a UAS platform, and plant data including plant height, number of leaves, crop and variety, planting date, stipulated herbicide application, spectral data and other information on soil and plants in the immediate vicinity, etc. These data will be used to establish a spectral signature for Palmer amaranth, and to develop a classification strategy to distinguish it from other plants in the surrounding area, including crops and other weeds.Study of the relationship between beef cattle temperament and beef quality using video, and thermal imaging technologiesFor every animal, a ten second video will be captured and saved in the same format. Each video will be divided into ten pictures (one picture per second) through image processing software. For each animal, a video processing method will be generated for each second of the video.Each image will be preprocessed in the same way and all of resulting images were subjected to extraction and analysis of the distance difference. The distance calucation algorithm is based on the pixel distance in the video frame between marker 1 and marker 2. All algorithms used in this study for video image preprocessing and analysis will be developed in MATLAB and NI LABVIEW by the PI's team.For each animal, thermal image mapping of the head will be acquired by a handheld NIRS camera (FLIR E8 Infrared Camera, FLIR Inc., OR). The thermal sensor measurement ranges from 4° F to 482° F (20° C to 250° C) with 2 percentage or 2° C accuracy. The computer software will be linked with image processing software to analyze the thermal image for each animal. As shown in Fig. 4, certain regions of interest (hot spots) of temperature will be measured and extracted through thermal imaging software for coloration analysis. Environmental factors like extreme low temperature (-20° C) will significantly affect our thermal sensor measurement result. All the animals will be used in this research will stay the standard sensor operating ranges (10° C -40° C) to minimize the errors from the environmental factors. The correlation between surface thermal value, beef cattle temperament and beef eating quality will be analyzed in the end of this research.Beef eating quality attributes (tenderness, color, pH) will be evaluated on the longissimus muscles in the North Dakota Meat Quality Lab. Steaks will be first evaluated for subjective color, intramuscular pH, then to be cooked to medium rare degree of doneness (65 °C), cooled, and evaluated for mechanical toughness using the Warner-Bratzler shear force method in accordance with the protocol published by the American Meat Science Association (AMSA, 1995). For statistics analysis, two procedures will be proceeded to study the relationship between beef quality and image processing methods. First, the correlation analysis method will be used to determine the useful image features to use as predictors in model development. Second, linear regression method will be used to establish the prediction model of beef cattle temperament measurement. The statistical analysis was performed using SPSS 20.0 (SPSS Inc., Chicago, Illinois, USA). Due to the reality of the experiment (limited funding for animal handling, feeding, etc.), this study will be superimposed on trials with cattle on feed at the NDSU, where any treatments imposed will not affect the outcomes from the imaging work.