Optimization of lighting parameters for edible bird's nest vision inspection system
Loading...
Date
2018-05
Authors
Gwee, Kai Li
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Edible bird’s nest production is an important food industry in South East Asia
with increasing demands owing to its medical benefits. The edible bird’s nest
manufacturer looking into mechanization to increase the productivity because the
existing cleaning process of the edible bird’s nest is labour intensive and time
consuming. Therefore, machine vision inspection system on the edible bird’s nest was
introduced but it is still in research. This is because, up to date, the details of a complete
and optimised edible bird’s nest vision inspection system, especially the system setup,
is still not stated in detailed. While the machine vision system setup that is being
practised widely in a various type of agriculture products is not suitable for the edible
bird’s nest inspection due to the complex, random and uneven structure of edible bird's
nest which greatly affect the transmittance and reflectance of the lighting during image
acquisition. Thus, this project is to optimize the lighting parameters for the edible bird’s
nest vision inspection system. The lighting parameters such as the type of lighting, the
angle of lighting, the wavelength of lighting and the intensity of lighting are explored
to obtain large contrast value between the impurity features and the edible bird’s nest
during the image acquisition. The optimal lighting parameters for the edible bird’s nest
vision inspection system is selected by using the full factorial design. The optimal
experimental setup is made up of the red front lighting that placed at 60˚ from the edible
bird's nest specimen with 255 intensity of lighting. The images acquired under the
experimental setup with the optimal lighting parameters are used for segmentation. The
adaptive threshold is used to segment the impurity features from the EBN which is then
compared to the impurity features that detected by a human expert. Besides that, a
convolutional neural network is trained for classifying the images according to its
cleanliness. Then, both proposed methods were compared with their performances. The
results show that the optimal edible bird’s nest vision inspection system successfully
achieved a detection rate of 84.44% with the false detection rate of 19.84% by using
adaptive threshold while the correct classification rate of the neural network is 76.47%.
As compared to previous works, this research shown an improvement in the impurity
features detection with the optimal lighting parameters.