Identifying carrot appearance quality by transfer learning

Abstract

Abstract It is extremely important to correctly identify the carrot appearance quality in design and manufacturing of Carrot sorter. In this paper, we have established a carrot appearance quality control system based on deep learning framework. The information of carrot is collected using the image, and thereafter the recognition model is erect on AlexNet network, which is pre‐trained by a large‐scale computer vision database (Image‐Net). Our framework uses transfer learning, which trains neural networks with small amounts of data compared to the traditional CNN. Applying this approach to the data set of carrot images, to demonstrate the classification performance of the proposed model, we compared it with human experts. The carrots of different grades can be recognized from a large number of carrots under the different surface condition with great accuracy. For binary classification recognition, the accuracy rate was 98.70%, the sensitivity was 98.34%, and the specificity was 98.99%. At the same time, when the number of training samples is small, high recognition accuracy can still be achieved through transfer learning. The model can not only meet the requirements of classification recognitionbut also greatly reduce the cost paid in sample collection. Practical Applications It is extremely important to correctly identify the carrot appearance quality in design and manufacturing of Carrot sorter. Using transfer learning, the training model can be constructed quickly, achieving high recognition accuracy and saving a lot of costs. Because the appearance of carrot plays an important role in quality. Therefore, this work has practical application value.