Research on the Recognition Method of Wheat Ears Based on Image Color and Classical Model

In this paper, we study wheat images during the grouting period, and construct a dataset by extracting wheat spike features through image preprocessing, color space conversion and edge detection. CNN and AlexNet models are used to train and optimize the parameters and structure to improve the recognition accuracy. By adjusting the batch size, learning rate and training rounds, CNN performs optimally with Batch size=64, Epoch=5, and learning rate=0.0001; AlexNet also performs well with similar settings, but the training cost is larger. This study verifies the effectiveness of image processing combined with the classical model CNN for recognizing wheat sheaves, providing data and theoretical support for wheat yield prediction.