Detection of maize drought based on texture and morphological features

The greatest impact on maize growth and yield at present is vegetation water stress. Therefore, a timely drought detection in maize is beneficial in arranging irrigation and ensuring the final return. Some methods use spectral reflection, infrared temperature measurement and chlorophyll fluorescence for drought detection. However, these types of equipment are bulky, incur high cost and cannot be widely used in an in-field environment. To alleviate these issues, we propose herein a method for detecting drought in maize from three aspects: colour, texture and plant morphology via computer vision. Compared to other methods, the average angle and dispersion of maize leaves are first calculated using a superpixel method. The morphological features of maize are then effectively described. Tamura and grey-level co-occurrence matrix is applied to extract the texture feature. Finally, we build a drought detection model using a support vector machine. Three water level datasets consisting of 1297 images is constructed to verify the method effectiveness. The final recognition rate is 98.97% by experiment, and it has good adaptability to light conditions in different periods of the day.