A calculation method for cotton phenotypic traits based on unmanned aerial vehicle LiDAR combined with a three-dimensional deep neural network

Accurate and rapid acquisition of cotton canopy phenotypic traits is relevant for growth monitoring, yield prediction, precise spraying of pesticides, and other scientific management. Manual measurements are timeconsuming and labor-intensive. Light detection and ranging (LiDAR) can accurately acquire point cloud data of the agricultural environment. However, LiDAR data require specific algorithms for processing and interpretation, making them unsuitable for direct use in agricultural applications. This study proposes a high-throughput detection method for phenotypic traits in cotton canopies based on an unmanned aerial vehicle (UAV) LiDAR platform. It comprises three key components: first, high-throughput data collection of field-grown cotton at the boll stage is conducted using the UAV LiDAR platform. Second, the three-dimensional deep neural network PointNet++ is used to process the raw data for semantic segmentation to extract cotton single-plant and block. Finally, six single-plant cotton phenotypic analysis algorithms and five block-level cotton phenotypic analysis algorithms are used to extract canopy structural information, such as cotton plant height, porosity, and canopy volume. In the final result, the extraction rate of the neural network for cotton single plants reached 86.3 %. Among the six methods for calculating cotton phenotypes, the plant height method was the most effective for calculating plant height, with an R2 value of 0.91 and the smallest root mean square error (RMSE) of 0.034 m, compared with the manually measured data. In the calculation result of cotton canopy porosity algorithms, the highest R2 value is 0.87 and the smallest RMSE value is 0.012. In the calculation result of cotton canopy volume algorithms, the highest achievable R2 value is 0.96 and the smallest RMSE value is 0.019 m3. The method can effectively partition cotton, extract phenotypic information, and provide technical support for cotton growth monitoring, yield prediction, and scientific management.