Investigation on the Recognition of Seeds Using Novel Color Features for Quality Seed Selection

Paddy is the chief and widely grown food crop in the world and is the staple food of more than 60
percent of the world population. A new feature set based on color was developed to detect paddy
seeds. This method has been applied to evaluate the power of various feature models based on six
color feature sets. Images of all seeds were acquired with an in-house imaging system and processed
with neural networks. The features applied as input for classification are the basic color features, and
the new features derived from these basic features. These features were employed to establish the
detection model for four paddy varieties viz. Karjat-6, Karjat-2, Ratnagiri-4 and Ratnagiri-24. The most
satisfactory feature from the color features was identified for accurate classification. Performance of
various feature models was investigated. Features based on method 3 were found poor, and method
6, where the normalized values are subtracted from all the three bands, showed better performance.
The performance was in majority for most of the instances of classification. This method 6 called as
proposed-color2 achieved an accuracy rate of 88.0%. Thus color moments have the potential to
improve the classification accuracy of the computer vision systems. The accuracy was less because
K6 and K2 have more or less same color so to increase the accuracy the shape and texture features
could also be included.