Multiscale Wheat Lodging Parameter Detection Based on MobilenetV3

Wheat is a major global food crop and plays a key role in food production and food supply. Wheat collapse has serious impacts on yield and quality, and it is crucial to obtain timely and accurate information on wheat collapse parameters. The aim of this study is to detect wheat lodging parameters using UAV images and to investigate the effect of UAV flight altitude on the model classification performance. Three UAV flight altitudes (15, 45, and 91 m) were selected to acquire images of the wheat test field, and an automatic segmentation algorithm was utilized to generate datasets at different altitudes. An improved model, MobilenetV3-M, is proposed to classify different wheat lodging parameters, and the model employs Pyramidal Convolution Pyconv (Pyramidal Convolution) to enhance the feature extraction capability of images at different scales and enhances the feature expression capability of the network through the introduction of the CA (Coordinate Attention) attention mechanism and combining it with PolyFL (PolyFocal Loss) to address the impact of data imbalance on model classification performance. The results show that the improved MobilenetV3-M has the best classification performance, with an average accuracy of 86.78%, 86.39%, and 84.35% in the datasets Angle, Location, and Ratio, respectively. Comparing the machine learning method SVM (Support Vector Machine) and the deep learning methods EfficientnetV2, MobilenetV2, and MobilenetV3, MobilenetV3 performs the best at all heights, with three datasets averaging accuracies of 76.70%, 78.17%, and 74.70%, respectively. With the increase in UAV flight altitude due to the loss of image feature information, the classification performance of both machine learning and deep learning models is affected, and this study provides a technical solution for accurately obtaining the fall parameters of wheat UAV images. The proposed improved model achieves high accuracy. It provides a new solution for wheat lodging warning and crop management, which is important for improving wheat production efficiency and sustainable agricultural development.