Abstract
Intercropping maize and soybean with distinct plant heights is a typical practice in diversified cropping systems, where shadows cast by taller maize plants onto soybean rows pose significant challenges for image based recognition. This study conducted experiments throughout the entire soybean-maize intercropping period to address illumination variation. Based on the height difference between crops, solar elevation angle, and light intensity at the top of the soybean canopy, an illumination compensation regression model was developed. The model was applied to correct soybean canopy images and compared against traditional enhancement methods, including histogram equalization, Multi-Scale Retinex (MSR), and gamma correction. Quantitative evaluation using peak signal-to-noise ratio (PSNR) showed the proposed method achieved 40.79 dB, indicating superior image quality. Furthermore, analysis of RGB and HLS channels revealed a consistent increase in brightness from left (darker) to right (brighter) across the images. Specifically, green channel values rose from 150-230 to 180-240, and overall RGB values exceeded 150, suggesting improved brightness and reduced local fluctuations. Brightness increased from 90-200 to 150-220, with the left region rising from 125 to 175. Finally, a comparison of channel-wise standard deviations among methods showed that the proposed algorithm exhibited lower variance in the green (G) and hue (H) channels, with favorable consistency across others. These results demonstrate the model's effectiveness in achieving smoother brightness transitions, thereby enhancing image uniformity and mitigating the negative impact of uneven illumination on recognition tasks.