| Rice yield is of great significance for national development planning and ensuring stable operation of the country.In recent years,many methods have been proposed to improve rice yield.The rice leaf age management model is a new method that first divides multiple growth and development process time points based on the external changes of rice,and then seizes these time points to carry out agronomic measures such as water and fertilizer management,in order to accurately regulate high-yield cultivation techniques and water and fertilizer management systems,and achieve the goal of increasing yield.Rice leaf age can more clearly reflect the phenological state of rice during the jointing and tillering stages.Therefore,the rice leaf age management model has developed into a cultivation technology system based on leaf age diagnosis,prediction,and regulation,which enables precise control of rice growth and development processes through the process of planting,managing,and harvesting according to rough time in traditional agricultural activities.In order to automatically and accurately determine rice leaf age in complex field environments,this paper proposes a rice leaf age recognition method based on semantic segmentation and shortest distance algorithm.The research content and results mainly include:(1)The semantic segmentation model called RLNet was constructed to segment rice leaf structure line regions.The semantic segmentation model is mainly based on the U-Net network,which is a structure that connects the encoder and decoder through skip connections.In order to further enhance the ability of network encoders to extract features,the encoder was replaced with ResNet34.ResNet34 is composed of many residual blocks,which prevent the learning ability of the network from decaying during the deepening process.Then an attention module was introduced,which can simultaneously learn channel attention and spatial attention.The module finally outputs a matrix of the same size as the feature map,in which each value represents the corresponding weight of each pixel.After multiplying it with the original feature map,the output of the encoder can be obtained.Finally,the input layer and the second to last layer are directly connected to further solve the problem of information loss during the sampling process on the feature map.By using this semantic segmentation model,the input rice leaves are divided into three rice leaf structure line regions.(2)Due to external factors such as shooting angle,lighting,and shadows,the segmentation effect of the RLNet semantic segmentation model cannot reach 100%.Therefore,there may be breakpoints in the segmented rice leaf structure line areas,and it is necessary to merge the disconnected areas.A coordinate based region merging algorithm is proposed to address this issue.The algorithm first filters the noise in the image,and then marks each remaining region,mainly in the form of coordinates.The algorithm marks the minimum ordinate,maximum ordinate,median ordinate,and average abscissa of each region.Arrange each region and traverse from top to bottom.When two regions meet the merging requirements,merge them.After traversing,each region is the complete merged region.Finally,by polynomial fitting,each region can be fitted into curves,which are called rice leaf structure curves,with an accuracy of 96.23%.(3)The shortest distance algorithm was proposed to determine the deviation of midvein in rice leaves and identify rice leaf age.Firstly,traverse all points on the middle leaf structure curve from top to bottom,taking one point as an example,calculate the distance from all points on the left leaf structure curve to this point,and take the shortest distance L from it;Next,calculate the distance from all points on the right leaf structure line to this point,and take the shortest distance R from it.Compare L and R.If L is shorter,increase the left count.Otherwise,increase the right count.Count the values on both sides.The larger the value on either side,the more biased the midrib is.Finally,by combining the length of the leaves,the rice leaf age can be identified,with an accuracy rate of95%.(4)The yield experiment was conducted in the experimental paddy field,dividing the field into an experimental group and a control group.The experimental group was managed using the rice leaf age management mode,while the control group was managed according to the commonly used field management mode.After harvesting,the average yield of two groups was statistically analyzed,and it was found that the average yield of the field using the rice leaf age management model was significantly improved,proving that this model is helpful in improving rice yield and also indicating that the leaf age recognition method proposed in this paper is meaningful.This study is based on the leaf vein bias method,demonstrating the feasibility of semantic segmentation technology for rice leaf age recognition.Unlike commonly used rice leaf age recognition methods,the method proposed in this article does not require complex manual assistance and harsh experimental conditions.It only requires taking photos of rice leaves in the field environment to automatically recognize rice leaf age,thereby reducing dependence on professionals and effectively improving rice yield.. |
