Research On Near-infrared Classification And Recognition Method Of Rice Blast Resistance Based On AW-1DCNN-XGBoos

Rice blast is a disease of rice caused by fungal explosions that occur in rice producing regions around the world.Rice can be damaged during the entire growth process from sowing to harvesting.In the years when rice blast was prevalent,the crop yield was generally reduced by 10%to 20%,and in severe cases,it was 40%to 50%.Some fields and even grains had no harvest,which caused great losses to the yield and quality of rice.The advantages of Chinese japonica rice are high rice yield,good viscosity after cooking,little swelling after absorbing water,but its resistance to rice blast is very weak.Cultivar resistance is the main factor affecting the control effect of rice blast.The key to control rice blast is to cultivate rice varieties with good resistance.Resistance screening of rice seeds is an important step before breeding,as most resistant varieties lose resistance 3 to 5 years after being popularized in the field.Near infrared spectroscopy(NIRS)has the characteristics of high efficiency,fast,non-destructive,and environmental protection.It is applied to qualitative and quantitative analysis of crops.With the wide application of near infrared spectroscopy in Chinese agriculture,near infrared spectroscopy has become one of the important methods for rapid detection of rice seed quality,which provides a new way for identification of rice blast resistance.Selecting high quality japonica rice varieties with blast resistance for planting and breeding,ensuring good yield,quality type,adaptability and stress resistance of seeds at the same time,improving the ability of rice to resist blast infection at the root,preventing the occurrence of rice blast,is of great significance for the improvement of rice yield and quality.The key research methods and conclusions of this paper are as follows:(1)This paper uses the improved deep learning model AW-1DCNN-XGBoost.The AW-1DCNN-XGBoost model and six models of two types were constructed using the full band data of the original spectrum:Deep learning models:One-dimensional Deep Convolutional Neural Network(1DCNN),AW-1DCNN,XGBoost;The traditional models are well Probabilistic Neural Network(BP),Support Vector Machine(SVM)and probabilistic Neural Network(PNN).The operating time was 21.44s,115.12s,40.07s,51.46s,869s,4s and 3s,and the accuracy rate was 99.38%,98.15%,98.15%,94.44%,100%,60%and 61%,respectively.The experimental results show that the running time of AW-1DCNn-XGBoost is much shorter than AW-1DCNN and XGBoost,and the accuracy is higher than AW-1DCNN and XGBoost.Although AW-1DCNN-XGBoost runs longer than SVM and BP,its accuracy is much higher than SVM and BP.Although AW-1DCNN-XGBoost’s accuracy is slightly lower than BP’s,its running time is much shorter than BP’s;Therefore,the improved deep learning algorithm AW-1DCNN-XGBoost has obtained satisfactory experimental results in the classification and identification of blast resistant rice seeds.(2)By analyzing the molecular structure of Pi-ta in rice blast resistance gene,we selected four characteristic bands containing-OH and-NH in the original near-infrared spectrum according to the absorption peak,and the first order differential spectrum contains five characteristic bands.We build a fast recognition model based on AW-1DCNN-XGBoost based on these 9 bands and their different combinations.The experimental results show that the combination of the original spectral band 6557.875cm-1～7262.25cm-1 and the first order differential spectral band 5107.879cm-1～5523.924cm-1 has the best effect,with a running time of 10.8s and an accuracy of 99.38%.Compared with full-band modeling,the AW-1DCNN-XGBoost recognition model established by using combined bands is twice as fast with the same accuracy.(3)Finally,the software of rice blast resistance classification and recognition system was designed and implemented.The front-end uses HTML,CSS,Javascript,Python script,and other technologies to achieve a visual user interface,and the back-end uses Python to build a classification and recognition model for rice blast resistance.The system functions include uploading near-infrared spectral data of rice seeds,displaying near-infrared spectral images of rice seeds,detecting rice blast resistance,and ultimately implementing a classification and recognition system for rice blast resistance.