Color contour texture based peanut classification using deep spread spectral features classification model for assortment identification

Abstract

Agriculture is one of great economic growth in India by producing variety of peanuts for various co products development. Peanut Crop and weed identification is an important step in developing a highly efficient automotive peanut production system. Most of the exiting classification fix the threshold margins to select the inappropriate features cause classification inaccuracy. Deep learning, as a powerful image processing technology, can extract conventional and characteristic features in layers. Spectral information is used as the input of the network, and the output spectral characteristics of the network are directly passed to the subsequent classifiers to achieve pixel-level classification. To resolve the classification categorization problem, we propose a Color contour texture based peanut classification using deep spread spectral features classification model for variety identification. Initial the Cascading subset Feature Filtering (CSFF) is applied to select the shape textures, color gradient rate form histogram equivalence and returns the feature scalar values. Based in CSFF the feature weight is attained to Social Spider Optimization to select the Impact Features Weight (SSO-IFW). The optimal object segmentation war applied based on cross fold object verification to select the shape projection features. Then Ada Boosting classifier generate the type of object structural for identifying type definition to generate pattern to produce peanut characteristic clusters. Based on Image Features recognition clusters, deep spread spectral neural classification is applied to train the features with soft-max logical activation function with Adaptive Convolution Neural Network (ACNN) to categorize the peanuts based on feature characteristics using image recognition, 20 varieties of peanut are collected through scanner type dataset to train the images into deep neural network. This produce high classification accuracy up to 89 %, high precision, recall rate under F-measure with low false rate. This achieves high classification than other existing methods.