Day-Old Chick Sexing Using Convolutional Neural Network (CNN) and Computer Vision

As the Research and Development (R&D) Center of Zamboanga Peninsula (ZAMPEN) native chicken in the Philippines, the center supplies ZAMPEN native chicken to growers within the peninsula and its nearby regions. Most growers prefer to buy day-old chicks because of less handling and transfer costs. However, the accuracy rate of R&D staff in classifying day-old chicks thru feather sexing has an average of 55%. Because of this, the center does not sell day-old chicks to growers. This research aims to develop a device that classifies the sex of a day-old ZAMPEN native chick using CNN and computer vision through feather sexing. Two hundred (200) feather images were collected, consisting of 100 females and 1000 males. And 20 no-detection images were added to the data sets. The model was created by sequentially adding layers using VGG-16 architecture. The prototype captures a video of the day-old chick's feathers. Then the video frame is cropped and processed for classification. Thirty (30) day-old chicks were used to test the prototype and got an 83.33% accuracy.