Abstract—Surface crack is a type of defect which depreciates dates quality. An automated system to detect and sort cracked dates is needed in the factories in order to ensure quality. The objective of this study was to determine the efficiency of a computer vision system with RGB color camera to classify dates based on surface cracks using Back Propagation Neural Network (BPNN). A total of 315 samples were imaged using a digital color camera. Each image was processed in MATLAB software and converted to HSV plane. The binary image of the segmented sample was used to extract features for the determination of surface cracks. Back propagation neural network (BPNN) was implemented to obtain the classification accuracies of the developed algorithm. The neural network classified dates into three classes (no-crack dates, low-crack dates and high-crack dates) with 77% accuracy. The accuracy was improved to 90% while classifying into two classes (without-crack dates and with-crack dates). The developed algorithm may be modified further and used to detect cracks on other dried fruits and vegetables.

Index Terms—BPNN, color imaging, dates, surface defect

Cite: Sawsana Al-Rahbi, Annamalai Manickavasagan, and Gabriel Thomas, "Back Propagation Neural Network (BPNN) to Detect Surface Crack on Dates Using RGB Images," Journal of Medical and Bioengineering, Vol. 4, No. 1, pp. 67-70, February 2015. Doi: 10.12720/jomb.4.1.67-70