Study On Soybean Plant Phenotype Measurement System Under Low Illumination Conditions

High quality and high yield of crops have always been the goal of agricultural production.How to improve the yield and quality of crops is the focus of intelligent agriculture.By sensing the growth status of crops,measuring the phenotypic information shown in the growth process of crops,realizing the prediction of crop growth trends,and adjusting crop planting measures in time according to their growth trends,it is conducive to improving the growth environment and nutrient supply of crops,and effectively improving the yield and quality of crops.Traditional crop information collection is mainly through manual field measurement,which is not only affected by subjective judgment,but also easy to cause irreparable damage to plants;it is difficult to obtain detailed phenotypic information by manual work,especially in low illumination environments such as night and cloudy days.Therefore,the traditional crop phenotypic information acquisition method can not only realize the monitoring of crop growth status at night and cloudy days,but also affect the timely adjustment of crop growth environment and nutrition supply.With the continuous development of computer vision technology,it is an inevitable trend to apply it to the perception of crop growth status to develop intelligent agriculture.In this paper,computer vision technology is used to measure the phenotype of soybean plants under low illumination conditions through three-dimensional model reconstruction and leaf segmentation.On this basis,a soybean plant phenotype measurement platform is built.Firstly,due to the insufficient brightness and detail loss of soybean plant images under low illumination conditions,the matching accuracy between images is low during the reconstruction process,and the estimated spatial point cloud and camera trajectory are inaccurate,which makes the reconstructed soybean plant model less accurate.The measured phenotypic parameters have a large deviation from the actual,and the plant growth status cannot be truly expressed.Aiming at this problem,a three-dimensional reconstruction method based on soybean plants under low illumination conditions is proposed.The improved Enlighten GAN network is used to enhance the collected low-illumination images,restore the detailed information of soybean plants,and highlight the structural characteristics of plants.The scale invariant feature transformation algorithm is used for feature detection and matching to obtain the correspondence between soybean plant feature information and multi-view.Combined with accurate point cloud information and stable camera pose,the point cloud model of soybean plant was reconstructed by using motion recovery structure algorithm and multi-view stereo algorithm based on patch.The experimental results show that the reconstructed plant model is close to the morphology and traits of real soybean plants,which is helpful to measure the overall phenotypic data of soybean plants.Secondly,in order to obtain comprehensive phenotypic information of soybean plants,while measuring the overall phenotypic data of soybean plants,the local phenotypic information of soybean plants is obtained by measuring the phenotypic data of soybean leaves,so as to achieve a more accurate evaluation of the growth status of soybean plants.Due to the complex background of soybean leaf image,there are problems of missed detection and false detection in the process of leaf segmentation.An instance segmentation method based on deep over-parametric convolution is proposed.Based on the Yolact Edge model,the deep over-parametric convolution is used to replace the ordinary convolution in the network to improve the feature extraction ability,and the convolution block attention is introduced for feature integration to improve the segmentation performance of plant leaves.The experimental results show that the accuracyAP_all of the mask Mask and the detection frame Bbox on the soybean leaf verification set is improved by 1.32%and 0.54%respectively.Finally,based on image enhancement,three-dimensional reconstruction,and instance segmentation models,a soybean plant phenotype measurement platform was designed and built to realize the measurement,storage,and visualization of comprehensive phenotypic data of soybean plants.The test results show that the measurement platform can quickly and accurately measure the phenotypic parameters of soybean plants and evaluate the plant growth status in time.Figure[62]table[6]reference[84]...