A non-destructive technique based on texture analysis for identification of defective alphonso mangoes using subspace analysis techniques.
Musale S. S., Patil P. M.
Author Affiliation: Sinhgad College of Engineering, Pune - 411 004, MS, India.
International Journal of Agriculture Sciences 6 : 388-392
Abstract : In many circumstances, texture is the only information that can be used in natural image analysis because it is an important property of the surface that characterizes it's nature. Texture is defined as a spatial arrangement of local (gray level) intensity attributes which are correlated within areas of visual scene corresponding to surface regions. An image region has a constant texture if sets of its local properties in that region are constant. Thus texture analysis has received considerable attention in the field of image analysis and pattern recognition. Texture exhibits some sort of periodicity of the basic pattern of Spongy Tissue in alphonso mango. This leads to use textural property to identify different patterns of Spongy Tissue in alphonso for detection of defects in alphonso mango. Visual assessment of texture made by human is time consuming and inspection made by human does not achieve a high degree of accuracy and preciseness. Automated visual inspection of the textural pattern improves the accuracy and preciseness during detection of defects in alphonso mango. In the literature, the researchers worldwide have been working in various texture analysis algorithms for different applications like detection, recognition, classification, segmentation, clustering etc. Many algorithms suffer from low sensitive detection, difficult back ground adaption and high memory requirement. Problems and limitations associated with the available techniques have been reported by many studies. Each has some draw-back under all lighting conditions and no one has used a robust, reliable algorithm for detection of spongy tissue in alphonso mango under real life test environment. To develop an optimized algorithm using a non contact mechanism which will detect the defective alphonso mangoes happen to be a challenging task. The objective of the proposed research work is to obtain computationally cost effective and noncontact solution that achieve better recognition rate under various conditions in consultation with the agriculture scientist. In this paper we have proposed use of subspace analysis techniques for extraction of textural features that identifies Spongy Tissue in alphonso mango successfully. Performance of the proposed algorithm is carried out on the alphonso database [1]. This paper presents a methodology that combines the principal component analysis (PCA) and fisher linear discriminant analysis (FLD) applied for defect detection in alphonso mangoes. Performance of the proposed algorithm has been compared with that of PCA and FLD applied individually for both healthy and defective alphonso mangoes. Experimental results computed using the proposed integrated algorithm, individual PCA and individual FLD have been validated manually with the cut sections of the alphonso mangoes available in the database. It is observed that the proposed method shows significant improvement for both defective as well as healthy alphonso mangoes over PCA and FLD applied individually. The feature vectors extracted by using integrated PCA-FLD method have 99% of the highest discriminant power.