Classification of pistachio with deep learning and sorting with robotic manipulator

As the industrialization grows, demand for automation increases with it. Quality assessment is an important step in production line, which also requires automation for it comes with many benefits such as continuous production. Objective of this study is to make an automated pistachio separation system. Today, quality inspection of pistachio is done manually which reaps the potential benefits of automation. To automate this step, a deep algorithm specialized in object detection has been trained. Algorithm was trained to classify the pistachios into four classes: red hulled pistachio, yellow hulled pistachio, cracked pistachio and not-cracked pistachio. The algorithm was trained to recognize pistachios when they were clustered together to have a more realistic assessment. The algorithm's detection rate on the recognition of the individual pistachios was near 100% and confusion matrix show very promising classification accuracy, with accuracies of 98% for red hulled pistachios, 94% for yellow pistachios, 74% for cracked pistachios and 89% for not-cracked pistachios. Further in the thesis, image segmentation is done to get the area, length and width of the pistachios. Tests shows an average of 1.77% error for the accuracy of measurement. In the last part of the thesis, a robotic manipulator was made to physically separate pistachios. Test shows that, grippers needs an upgrade because it is unable to separate pistachios in one try when they are clustered together, but when they are separated from each other for a small distance, separation was observed to be possible.