UCSF

Presently, the quality of cherries in the market is uneven, because human senses are used to distinguish cherry quality, which consumes a lot of time and energy and does not achieve good results in terms of accuracy. If the internal quality indices, such as the PH value, sugar-acid ratio, and vitamin C content, of cherries are extracted using chemical methods, the detection speed will decrease. With the development of artificial intelligence (AI), image processing by AI algorithms has attracted broad attention. The YOLOv5 model in the YOLO series has many advantages, such as high detection accuracy, fast speed, small size, and so on, and has been used in face recognition, image recognition and other fields. However, owing to the influence of seasonal weather, the environment and other factors, the dataset used in the training model decreases the accuracy of image recognition. To improve the accuracy, a large amount of data must be used for model training, but this will decrease the model training speed. Because it is impossible to use all data in training, there will inevitably be recognition errors in the detection process. In this study, the cherry images in a dataset were extracted by the flooding filling algorithm. The extracted cherry images were used as a new dataset for training and recognition, and the results were compared to those obtained with non-extracted images. The dataset generated by the flooding filling algorithm was used for model training. After 20 training epochs, the accuracy rate reached 99.6%. Without using the algorithm to extract images, the accuracy rate was only 78.6% after 300 training epochs.