Deep Learning Based Serine Protease Enzyme Dose Detector to Degrade Swiftlet Feathers in The Edible Bird Nest Verianti Liana (a*), Riris Waladatun Nafiah (a), Rizal Arifiandika (b), Bagas Rohmatulloh (b), Yusuf Hendrawan (b), Tunjung Mahatmanto (c), Dimas Firmanda Al Riza (b)
a) Department of Agroindustrial Technology, Universitas Brawijaya, Jl.Veteran, Malang, ZIP 65145, Indonesia
*veriantiliana[at]student.ub.ac.id
b) Department of Agricultural Engineering, Universitas Brawijaya, Jl.Veteran, Malang, ZIP 65145, Indonesia
c) Department of Agricultural Product Technology, Universitas Brawijaya, Jl.Veteran, Malang, ZIP 65145, Indonesia
Abstract
Edible Bird Nest (EBN) is a product of solidified swiflet saliva (Aerodramus fuciphagus) which renowed as Caviar of the East due to its expensive price. The main factors that determine the quality and price of EBN are color and cleanliness of EBN from the attached swiftlet feathers. The removal of swiftlet feathers in EBN is usually done manually which is destructive to the structure of EBN, reduce the quality and price of EBN, inefficient in terms of time and cost, and ineffective due to human error. The use of serine protease enzyme can degrade swiftlet feathers in EBN non-destructively, environmentally friendly, safe for food, efficient in cost and time, and increase the production of clean EBN. The serine protease enzyme needs to be applied accurately with different doses according to the weight of swiftlet feathers in order to degrade all swiftlet feathers in EBN effectively and efficient in enzyme costs to optimize the profit of clean EBN production. The purpose of this study was to detect the dose of serine protease enzyme in degrading all feathers in EBN according to the uncleaned EBN images using Convolutional Neural Network (CNN). CNN as the main deep learning architecture was used in this study because of its ability to learn image characteristics automatically with high accuracy in the image recognition process. The training accuracy with GoogleNet as the pretrained network of CNN in this study reached 95.35%.
