Fruit (Orange) Quality Inspection using Fully Convolutional Network

The quality assessment of fruits is a critical task in the agriculture and food processing industries, influencing consumer satisfaction and market competitiveness.Over the years, various methodologies have been explored to address this challenge. These include traditional color and texture analysis, machine learning algorithms, and convolutional neural networks (CNNs). While these methods have their merits, they often fall short in terms of accuracy and efficiency, particularly in detecting subtle defects.Our proposed system combines aspects of color and texture analysis to conduct an initial quality assessment. It utilizes machine learning algorithms for extracting features and incorporates deep learning techniques, specifically employing Fully Convolutional Networks (FCN) with a UNet architecture, to achieve accurate defect segmentation.By utilising the contrast between sound and defective regions, the suggested approach improves segmentation accuracy and is especially good at recognising pixels near defect boundaries on the orange surface. Background pixel removal, local segmentation-based picture binarization, image subtraction, morphological modification, and stem end pixel removal are all steps in the comprehensive image processing process.Additionally, the article suggests an automated grading system in recognition of the difficulties with manual sorting for the export market. The system categorises oranges according to ripeness and size using pattern recognition techniques on single-color photos of oranges.