The Chicken’s legs Size Classification using Image Processing and Deep Neural Network

This paper presents the technique to classify the size of chicken's legs using digital image processing (DIP) and deep neural network (DNN). The single camera has been applied to acquire the image. After that, the chicken's legs are segmented from the background using background subtraction, and its contour can be determined. The contour is applied to compute the contour's area and bounding rectangle with its minimum area, producing the width and length. The three information of contour are passed to a DNN constructed by rectified linear unit (ReLU) and softmax layer. It is trained and generated the classifying model. The target data are divided into three groups (Small, Medium and Large), which is based on the chicken's legs weight obtained by the standard weighing scale. The proposed technique is implemented on Python 3.5.5 supported by OpenCV 3.1.0 and Keras version 2.1.6. The experimental results show that the measurement error of object dimension is lower than ±0.2 cm and ±0.4 cm
²
for length and area, respectively. Finally, the trained model for classifying the chicken's legs size yielded 100% accuracy.