Enhancement of grading system for fruits using effective classification techniques

Abstract

Advances in Image processing and Pattern recognition have led to worldwide innovation in the field of agriculture. The need for this technique paved way for various applications like weed detection, vegetation measurement, irrigation, disease identification in plants, grading and sorting of fruits and vegetables are considered as the most challenging tasks in agriculture. The human decision of grading of fruit is not accurate all the time. As the concept of image processing techniques, fruit grading has been focused on this research. Conventional classification methods presented in various literatures apply segmentation and classification methodologies to solve the above mentioned problems. Segmentation is an existing method to provide quality information about the fruits. Classification is a more effective method to crop efficient grading. This thesis presents an enhancement of the mango fruit grading system. Many image processing approaches have been presented to solve the problem of fruit grading, but researchers have suffered with false classification rate and less accuracy. To improve the classification rate, MultiLevel Feature Distribution (MLFD) measure based approach for mango classification and grading, an efficient Genetic Adaptive Neuro-Fuzzy Inference System (GANFIS) and Multi-level Intensity Estimation Technique (MLIE) based fruit grading using template matching and fuzzy rule sets are implemented. MLFD presents fruit grading system with pre-processing of fruit image followed by Segmentation using K-means clustering algorithm to identify background and objects. The color Histogram method extracts color features from all the images. For the extracted color features, the processgenerates the histogram. Then, using the generated histogram values, the method computes the feature distribution matrix. After that for the input test image, the color histogram is extracted and the feature distribution measure is estimated and compare with the trained values. The input image is graded depending on the feature distribution measure. The pre-processed image from GANFIS approach is segmented using K-means algorithm which identifies defective and non-defective fruits. For non-defective mango image, Gray Level Co-Occurrence Matrix (GLCM) and Local binary pattern were applied to extract various feature parameters like color and texture features from the fruit image. Over the extracted features, the genetic algorithm is applied to execute the feature selection. The adaptive Neuro-fuzzy inference technique is applied to the extracted features to execute the mango classification and grading. This classification algorithm estimates the multi feature class similarity measure towards each class of mangoes to execute classification where the grading is performed based on the same being estimated within the clas newline