Accurate observation of the anatomical structure of the root system of aquatic plants is of great significance in understanding their adaptive mechanisms. In this paper, an image processing algorithm based on improved Retinex theory is proposed to address the problem of poor image quality and blurred details of underwater reed root system. The method combines the light-directed variable attention module (LDAT), the spatial-frequency domain feature fusion module (SDFF), and the medium transport module (MTM), and realizes the effective enhancement of the underwater image by introducing a perturbation term to simulate the underwater imaging environment and using saturation to distinguish between the artificial illumination and the natural light source. The experimental results show that compared with the traditional dark channel a priori algorithm and MSR algorithm, the algorithm in this paper improves the information entropy significantly, in which the information entropy of image 3 is improved from 6.762 to 7.827, and the standard deviation is increased from 18.978 to 58.973, and the average running time is only 6.696 s, which is significantly better than that of the dark channel a priori algorithm, which is 71.49 s. The anatomical structure of the root system is analyzed for the reeds under the different water depth conditions. By analyzing the anatomical structure of the root system under different water depth conditions, it was found that the proportion of root cortex aerated tissues was significantly positively correlated with the water depth, and the proportion of aerated tissues was significantly larger under deep flooding than non-flooding conditions. The conclusion shows that the algorithm can effectively improve the quality of underwater images and provide technical support for the accurate analysis of plant root structure.
