MORPHOMETRY AND CLASSIFICATION OF CHICKEN WHITE EGGS USING COMPUTATIONAL VISION AND ARTIFICIAL NEURAL NETWORK TECHNIQUES

This paper defines a protocol for geometric description and classification of white eggs of chickens using computer vision (CV) and artificial neural network (ANN) techniques. For this, 60 eggs were selected and their dimensions, volumes and surface areas measured using pachymeter, immersion method and mathematical expressions available in the literature, respectively, to compose the reference values. Subsequently, the eggs were imaged with RGB camera and the digital images analyzed using dedicated algorithm. The estimated geometric parameters were compared with the witness values, using descriptive statistics and t test, and used in ANN training for automatic classification of eggs. Perceptron was the algorithm adopted, different configurations were evaluated in the ANN and 20 digital images of white eggs available in the literature used in the classification validation. Prediction of the height, width, height/width, volume and surface area of the eggs using CV showed high correlation and low average square error with the reference values. The best egg classification results occurred with an ANN with three neurons and data of height, width and surface area as input parameters. The proposed protocol has proved feasible for the geometric description and classification of white eggs of Brazilian agro-industries.