Frost Amount Measurement Based On Imaging And Capacitive Characteristics Of Frosty Crop Leaves

Frost is a common weather phenomenon,and the frequent frosts in early spring across subtropical or temperate regions cause serious damage to crop production.At present,there is a lack of meteorological observation equipment for frost monitoring at home and abroad,and there are no techniques and instruments for crop frost detection,mostly relying on manual frost observation.In recent years,mechanized frost control techniques and equipment have been gradually promoted and applied in agriculture,but their control is based on a single temperature or inverse temperature difference as an empirical judgment condition,which could not achieve timely and accurate prevention and control of frost damage due to the lack of real-time detection/monitoring of frost parameters,thus resulting in a low level of automation and intelligence in practical application.This paper takes the whole process of crop leaf frosting as the research object,and discusses to determine the technical solutions of frost detection at the beginning,middle and after stages through model analysis and experimental observation respectively;experiments and establishes the corresponding frost identification and frost estimation models based on the different patterns of frost crystals/frost layers on leaves in each stage and validates them based on their different response characteristics of image and electrostatic capacity.The development of this key technology of frost volume detection is of great theoretical value and practical significance in the technical fields of agricultural frost protection and meteorological observation.The main research contents and results are as follows:（1）Analysis of crop leaf frosting process and frost amount detection technology methodFirstly,the morphological characteristics of frost crystals/layers were obtained by super depth-of-field microscopy,and the growth change characteristics of frost crystals/layers were analyzed;secondly,based on the condensation frosting process of discontinuous phase change and Eulerian multiphase flow control equation,the numerical model of frosting and frost quantity on crop leaf surface was established,and the effects of leaf temperature,air velocity and relative humidity on the thickness and quality of frost layer were analyzed.The effects of leaf temperature,air velocity and relative humidity on frost thickness and quality were analyzed.The results show that（1）there are three distributions with typical characteristics,such as sporadic,scattered and stacked,during the whole process of real-time frosting;（2）the numerically calculated frost volume increases with decreasing leaf temperature but decreases with increasing airflow velocity,and increasing relative humidity could promote the increase of frost-forming quality but may reduce the frost thickness;the deviation of calculated frost thickness is-6.25%～+11.76%,（calculated）mass deviation was-8.4%to+23.8%.Based on the microscopic morphological characteristics of frost crystals/layers on crop leaves and the results of numerical simulation of frosting process,the qualitative or quantitative detection techniques for frosting at the beginning,middle and end of frosting were determined,including:critical frost crystal detection based on microscopic imaging,frost area percentage detection based on image gradient enhancement,and frost volume estimation model based on electrostatic capacity response of frosted leaves.（2）Microscopic imaging-based detection of critical frost crystals at the early stage of frostingA microscopic imaging-based critical frost crystal detection（recognition）model,Frost-YOLOv5,was developed for early detection of very small（first（or several）grains or clusters）frost crystals at the early stage of frosting on crop leaves.Then,for the detection target of fine critical frost crystals,a critical frost crystal recognition model Frost-YOLOv5is constructed by introducing both channel and spatial attention mechanisms in the YOLOv5base model.the results show that the detection accuracy P of this model is 98.89%,the average accuracy AP is 81.72%,and Recall is 81.51%,where P is higher than YOLOv3-Efficientnet-b2,YOLOv4-Tiny-ECA,and YOLOv5 models by 3.69%,1.42%,and 2.31%,respectively.In the detection of frost crystals for two different target size types,the overall false detection rate of Frost-YOLOv5 is 3.14%,which is 4.4%lower compared with the YOLOv5 model before improvement,and in the detection of small target frost crystals,the false detection rate is reduced by 3.16%from the original 11.90%,indicating that the introduction of the attention mechanism is beneficial to the detection and identification of small target critical frost crystals.（3）Experimental study of mid-term leaf frost detection based on image gradient enhancement methodIn the middle stage of frosting where frost crystals are gradually dispersed on the leaf surface,it is suitable to characterize the amount of frost by the area ratio of frost crystals to the leaf as an indicator.For the three backgrounds of frosted leaf images,such as vignetting,soil and miscellaneous leaves,the gradient enhancement of HSV color space and edges is used to segment the leaf and frost crystals,and then the above area ratio is calculated and error analysis is performed.The results show that:（i）the average error rate R_ms in the three backgrounds of defocus,soil and miscellaneous leaves are 2.03%,3.71%and 4.12%,which are 5.34%,5.69%and 8.68%lower than those of Exg super green algorithm,OTSU algorithm and improved former HSV algorithm,respectively;（ii）among the three frost formation status classes,the frost formation area ratio on the leaf surface calculated based on the algorithm of this paper The results are closest to the reference value,and the overall average accuracy R_a is 97.70%,which is 10.05%and 9.58%higher than the improved former HSV algorithm and Exg super green algorithm,respectively,indicating that its overall segmentation effect is the best,and it could realize the calculation of frost-forming area percentage of leaf surface under various scenarios with good segmentation effect more accurately.As the area percentage gradually increases,the segmentation error of the algorithm in this chapter reaches a minimum of 0.40%,indicating a more accurate calculation in the case of medium to high area percentage of frost.（4）Capacity characteristics of frosted leaves and frost amount prediction model in the late frost stageThe amount of frost could be measured based on the electrostatic capacity response characteristics of leaves with different frost amounts.A device for measuring the electrostatic capacity of frosted crop leaves was developed using the Texas Instruments evaluation board integration,and the relationship between frost volume and electrostatic capacity was experimentally modeled and validated.The results show that the average relative error of the trials with leaf frost amount was 7.78%,and the error range was between4.30%and 11.27%;the average relative error of the trials with net frost amount was 5.71%,and the error range was between 1.05%and 17.69%,the prediction error of the regression model between net frost amount and capacitance was lower and better overall.The thesis proposes different detection indicators for the whole process of crop leaf frost and develops corresponding detection techniques,forming models and detection systems for critical frost crystal identification,frost area percentage and frost amount estimation at the emerging,middle and end of frost respectively,which not only enriches the theoretical system in the technical fields of agricultural frost monitoring and frost meteorological observation,but also has important practical significance for improving the intelligence of agricultural frost control.