Skip to main content
ARS Home » Plains Area » Mandan, North Dakota » Northern Great Plains Research Laboratory » Research » Publications at this Location » Publication #350632
Title: Sunflower floral dimension measurements using digital image processing

Author
item SHAJAHAN, SUNOJ - North Dakota State University
item NAVANEETH, SUBHASHREE - North Dakota State University
item BABU, DHARANI - North Dakota State University
item CANNAYEN, IGATHINANTHANE - North Dakota State University
item Franco, Jose
item MALLINGER, RACHEL - University Of Florida
item Prasifka, Jarrad
item Archer, David

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/10/2018
Publication Date: 6/1/2018
Citation: Shajahan, S., Navaneeth, S., Babu, D., Cannayen, I., Franco Jr, J.G., Mallinger, R.E., Prasifka, J.R., Archer, D.W. 2018. Sunflower floral dimension measurements using digital image processing. Computers and Electronics in Agriculture. 151:403-415. https://doi.org/10.1016/j.compag.2018.06.026.
DOI: https://doi.org/10.1016/j.compag.2018.06.026

Interpretive Summary: Measurements of the size of sunflower heads are important for estimating seed yield. Also, sunflowers depend on bees to move pollen between male and female plants to produce hybrid seed, and the size of flowers (floral display) can be important in attracting pollinators. Measuring the sunflower head, disc, and ray florets (petals) by hand can be subjective and time-consuming. Using images from a digital camera and adapting a computer program to measure sunflower dimensions can save time and give more accurate measurements. Image processing is becoming more common in agriculture, and can also be extended to pollination research. This study developed a method to get accurate measurements of sunflower head area using image processing. Not only does the method give more accurate results, this method can decrease the time it takes to measure sunflower floral dimensions in the field. The results can help reduce time in the field for researchers, but also potentially be used as a tool for growers to estimate seed yields.

Technical Abstract: Sunflower floral dimensions are essential for assessing pollinator attraction and estimating seed yields. Dimensions measured manually at present are subjective and time-intensive; therefore an image processing method was developed as an alternative, which was objective, non-destructive, produces multiple results, and rapid. A digital single lens reflex camera captured the images in the field using a custom-designed background board that had markers spaced at known distances for calibration. An ImageJ user-coded plugin was developed to measure the dimensions of individual sunflower components, such as head, disc, and ray florets. Two measurement methods, direct (using the binary image) and wrapping-polygon (using a polygonal enclosure) were tested. The ‘pixel-march’ method made multiple radial dimension measurements (diameter) on the segmented ray florets’ binary image, where the dimensions of all components were measured in a single computation. The effect of multiple measurements (2, 4, 8, 16, 32, 64, 128 and 180 along 0-180° angles) was studied to determine an effective number of measurements. Measurements of the developed method were statistically compared with ImageJ’s various standard output parameters, allowing for the development of user-friendly linear models with good performance. From the results, we observed that (i) a minimum of 32 measurements was necessary for measuring the sunflower head and ray florets dimensions, but only eight measurements were necessary for measuring the sunflower disc; (ii) wrapping-polygon method was efficient as it produced minimum absolute deviation, and performed ‘tip-to-tip’ measurement; (iii) equivalent diameter (ED) and fitted ellipse minor axis (MinA) were well correlated (r = 0.88) to the accurate mean 180 measurements (D180) for all sunflower components; (iv) linear models for predicting D180 using ED and MinA performed better (R-squared > 0.99) for head and disc than for ray florets (R-squared > 0.76); (v) user-friendly linear models using the mean of two manual measurements of the head (D2) for predicting D180 and area were good only for the head (R-squared > 0.92), and were not suitable for the disc (R-squared between 0.55 and 0.62) and ray florets (R-squared between 0.05 and 0.43); and (vi) developed image processing method results were accurate, quick (˜11 s), and have the potential to be adapted to other species.