Author(s): Zhuoyue Zhou; Zhiguo Pang

Linked Author(s):

Keywords: Image recognition; Deep learning; Convolutional neural networks

Abstract: Deep learning technology can extract and recognize the features of crop images, so as to liberate labor and improve work efficiency. Rape is the oil crop with the largest planting area and the highest yield in China, and its growth cycle can be roughly divided into seedling stage, flowering stage and maturity stage. This paper constructs image data sets of rape seedling stage, flowering stage and maturity stage by using expansion methods, and the datasets are randomly divided into training set, verification set, and test set at a ratio of 6:2:2. Based on the Anaconda integrated environment and taking loss function value, recognition accuracy and training time as evaluation indexes, the optimal parameter configurations such as the optimal learning rate, batch size and iteration times were obtained through multiple comparison experiments. Combined with the training curve, the experimental results of different convolutional neural network models VGG16, VGG19, ResNet34, ResNet50, ResNet101 and GoogleNet were analyzed. The evaluation indicators were divided into multiple intervals, and the values of different intervals were assigned for different scores, the scores of each indicator are added according to their corresponding weights to obtain the total score. The results show that the optimal model is GoogleNet.

DOI: https://doi.org/10.64697/HIC2024_P434

Year: 2024