ANT COLONY TECHNIQUE FOR OPTIMIZING THE ORDER OF CASCADED SVM CLASSIFIER FOR SUNFLOWER SEED CLASSIFICATION

Abstract

In Seed Testing Laboratory (STL), test for other distinguishable variety (ODV) is carried out for the seed sample received from the seed producer after harvesting. This is done to obtain foundation/certified tag from the seed certification department before marketing the product. This is currently done manually by observing the morphological characteristics of the seeds through the naked eye. This is a time-consuming process for the STL and there is a chance for human error in identifying ODV. Hence, there is the need for the machine vision technique to automate the process of identifying ODV in the seed testing laboratory. Various techniques for varietal identification of sunflower seeds are explored in this paper. Experiments are performed on a dataset that contains ten varieties of sunflower seeds and the success rate achieved using various techniques is reported in this paper. In cascaded support vector machine (SVM), the order in which the classifier blocks are arranged plays an important role in improving the classification rate. The main contribution of this paper lies in the manipulation of ant colony optimization technique for obtaining the order of cascaded SVM by maximizing the total probability of correct decision. When the SVM is cascaded in optimum order, classification rate has been increased from 88.32% (obtained using the actual order) to 98.82% for kernel linear discriminant analysis based boundary descriptors. The closed form expression for computing the total probability of correct detection of the constructed cascaded SVM classifier is also reported in this paper.

Existing System

Adaptive Red and Blue chromatic map (ARB), Support Vector Machine (SVM) classifier

Proposed System

In this work, geometrical features like boundary descriptors and texture features such as Fourier and Cosine descriptors are attempted. The most commonly used dimension reduction techniques such as Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), Kernel− LDA & PCA followed by LDA are attempted in this method. Multiclass SVM is modelled by the cascade of several binary SVMs. A metaheuristic technique based on ant colony optimization is proposed to optimize the order in which the blocks are cascaded by maximizing the total probability of correct decision. The total probability of correct decision is obtained using the probability of miss and probability of false alarm for the individual classifiers.

Architecture

BLOCK DIAGRAM