Bit plane slicing technique to classify date varieties

G.Thomas, M. Annamalai, G. Karuppannan, “Bit plane slicing technique to classify date varieties,” Agricultural Engineering International: CIGR Journal, Vol 17, No 2, June 2015.

Abstract: Varietal purity is an important parameter in the quality standards of dates. In general, variety identification is done by visual inspection method in grading and handling facilities. Online variety assessment using computer vision methods with minimum features and fast image processing and classification algorithms would be highly beneficial for the date industry. Three date varieties (Khalas, Fard and Madina) were classified using a single type of feature, Euler number, used on the eight bit planes available from gray scale images. An overall classification accuracy of 91.48% was achieved using a two layer neural network classifier with hyperbolic tangent sigmoid transfer function. Additionally, image segmentation was performed using the two most significant bit planes. Therefore, a complete feature extraction module based on logic values and morphological image processing as proposed here can be easily implemented in hardware.

• Introduction

• Although annual production of dates in Oman is 255,891 Mt, only 9000 Mt is exported which is 2.5 to 3.5%.

• Varietal purity, color, uniformity of size and absence of defects are some of the important quality parameters for dates in domestic and international markets.

• Materials and methods

• Samples for three date varieties (Khalas, Fard and Madina)

• 100 date images were taken for each variety

• Bit plane slicing and segmentation

• Euler number as classification feature

• The Euler number is a measure of the topology of an image.

• Using bit plane f_bp6

Classification using a neural network

Mean squared error of training and test data set while using s=0.3 (no segmentation) in a neural network

Conclusions

The use of bit plane slicing and the Euler number defined a methodology for image feature extraction that was used for the classification of three varieties of dates. The mathematical simplicity of the proposed approach can allow the implementation of such a signal processing module in hardware that only supports integer value operations. Furthermore, bit plane slicing was used for segmentation purposes as well. In order to validate the quality of the features, neural networks were used, and that obtained correct classification rates of more than 90%.