| China is a large agricultural country,and its potato planting area and output have always ranked first in the world.The phenotypic characteristics of potatoes are also an important reference for potato breeding,processing,production,storage,and sales.However,machine vision and near-infrared spectroscopy have been used to detect phenotypic characteristics of potatoes in China,and even manual detection methods are still used in some regions.Low efficiency,time-consuming,and high cost have also become a major difficulty in non-destructive testing of potato phenotypic characteristics.In recent years,with the rapid development of image recognition technology,neural networks have played a crucial role in the non-destructive detection of potato phenotypic features,which has greatly improved the accuracy and objectivity of detection.Therefore,based on image recognition and neural network,a non-destructive detection system for potato phenotypic characteristics was designed with an accuracy rate of more than 90%,which can intuitively indicate the potato phenotypic characteristics.The main research contents of this article are:1.The length,width and height of potato based on image recognition were studied.The HSV method was used to extract the contours of potato images,and the median filtering method was used to remove noise from the potato images.The area method was used to determine the length,width,and height of the potato according to the calibration object.The potato was successfully clas sified into large,medium and small,and the recognition accuracy reached 90%.2.Research on potato damage recognition based on Google inceptionV3 neural network.The potato epidermal texture features were extracted,and after processing and training the obtained features,a Google inceptionV3 neural network potato damage model was established,and the accuracy of damage recognition was 93%.3.Research on potato bud pit recognition based on Google inceptionV3 neural network.The pit features of the potato epidermis were extracted and repeatedly trained to establish a Google inceptionV3 neural network potato pit recognition model.The accuracy of the pit recognition was 94.5%. |
