Development and validation of an automatic and intelligent system for medullation and average diameter evaluation to alpaca, llama and mohair fibers

The aim of this work was to develop and validate an automatic and intelligent system (based in artificial intelligence) capable of quantify and identify fibers, by type of medullation in 5 categories. This work was carried out in Lima, Peru. To develop the software a trained model was generated based on You Only Look at Once for medullation and DenseNet for average fiber diameter (AFD), using Python. Language C was used to develop the graphical user interface. For the hardware; mechanical, electronic and optical subsystems was design and development. Samples of white alpaca, llama and mohair fibers (2108, 1858, 901 fibers, respectively) were evaluated for identification the fibers medullation. Additionally, AFD of 197 samples of white alpacas were measured with two methods. This system identifies 5 types of medullation and measures the diameter of the fibers. Each sample is evaluated in 40 sec, considering about 1500 fibers/sample. At two-proportion z-test of different fiber medullation types obtained with direct counting and our system no significant differences were found. At t-test of AFD obtained with OFDA device and our system no significant difference were found. The relationship between these methods was very stronger (r=0.95). The use of this system is recommended for fiber evaluation for the purpose of genetic improvement of fibers in animal production; purchase-sale, and processing of fiber to verify the quality of fibers; and research on medullation to increase knowledge about alpaca, llama and mohair fibers.