Apple classification based on surface bruises using image processing and neural networks

摘要

Maintaining prime fruit quality is the key to success in the fresh fruit business. Quality defects such as bruises in apples adversely affect their market value. Line-scan x-ray imaging has shown potential for detecting these quality defects. Quality assessment of apples with computer vision techniques is possible; however, two basic issues must be addressed before an automatic sorting system can be developed: (1) which image features best correlate with the fruit quality, and (2) which classifier should be used for optimal classification. These issues are discussed in this article. Red delicious (RD) and golden delicious (GD) apples were line-scanned for bruise damage. Spatial and transform features were evaluated for their discriminating contributions to fruit classification based on bruise defects. Stepwise discriminant analysis was used for selecting the salient features. Spatial edge features detected using Robert's edge detector combined with the selected discrete cosine transform (DCT) coefficients proved to be good indicators of old (one month) bruises. Separate artificial neural network (ANN) classifiers were developed for old (one month) and new (24 hour) bruises. When an ANN classifier was used to sort apples based on old bruises, it achieved an accuracy of 90% for RD apples and 83% (93% after threshold adjustment) for GD apples. For new bruises, the accuracy was approximately 60% for both RD and GD apples. New bruises were not adequately separated using this methodology.