Fruit Quality Assessment Using Texture and Shape Features

Abstract

Automated evaluation of fruit quality is an important part of the modern precision agriculture system in India. It helps reduce the wastage of crops. Traditional manual methods are often inconsistent and labor-intensive. In this paper, a unified computational framework is proposed for fruit quality classification based on multi-scale textural and invariant geometric features. The method applies Discrete Wavelet Transform (DWT) to extract the high-frequency energy components of the fruit surfaces. This helps in identifying textural defects like bruising and rot. At the same time, we extract shape descriptors (area, solidity, a set of seven Hu invariant moments) to generate a strong structural profile which is rotation and scale invariant. To demonstrate the efficiency of the selected descriptors, we performed a feature discriminative analysis, using distribution validation and correlation matrices. This confirmed that there was a significant statistical difference between fresh and spoiled fruit types. The extracted features were assessed using different machine learning algorithms such as Random Forest, Support Vector Machine (SVM) and K-Nearest Neighbor (KNN). Experimental results on a new dataset of fresh and rotten fruits such as apples, bananas and oranges, show that our combination of texture and shape features outperforms approaches that only consider a single type of feature by a large margin. The accuracy given by Random Forest classifier is 70.75 %. This demonstrates the efficiency and effectiveness of the fusion of the geometric solidity and DWT based energy as a solution to fruit classification system in real-time.