Research On Sugarcane Field Recognition Method Based On Support Vector Machine

Sugarcane is an important type of cash crop and plays a crucial role in global sugar production.Clarifying the magnitude of sugarcane planting will likely provide very evident supports for local land use management and policy-making.However,sugarcane growth environment in complex landscapes with frequent rainy weather conditions poses many challenges for its rapid mapping.Among them,the cloud noise in the sugarcane planting area will have a serious adverse impact on image interpretation.It is of great significance to explore a rapid and efficient remote sensing data cloud removal method and apply it to the recognition of sugarcane filed,which also puts forward requirements for images with high temporal and spatial resolution.This study attempted to map sugarcane filed using sentinel-2 remote sensing image with spatial resolution of 10 meters and phenological information of typical crops.Firstly,to minimize the influences of cloudy and rainy conditions,this study designed a corresponding threshold segmentation algorithm for six cloud-related bands/products of Sentinel-2 satellite data based on the GEE(Google Earth Engine)platform.A comparative study on cloud detection was conducted in a typical area of the study area of LongZhou county to explore the cloud detection potential of different bands/products and to screen out bands/products with strong sensitivity to cloud noise;Next,the best cloud detection algorithm is used to detect all images in the study area and cut out the cloud noise.Then,based on the idea of substitution method,all images of three typical phenological periods(seedling stage,elogation stage and harvest stage)in the study were fused to obtain Sentinel-2 cloud-free remote sensing images of three typical phenological periods in LongZhou county,Guangxi in 2018.Then,the fusion images were used to calculate the NDVI(Normalized Difference Vegetation Index)of each phenological period to obtain three NDVI spectral feature maps,and the three NDVI feature maps were synthesized to obtain a three-band NDVI image with phenological characteristics.Finally,based on support vector machine algorithm of polynomial kernel function,the three-band NDVI dataset along with 4000 training samples and 2000 random validation samples was used for sugarcane field recognition and mapping.To assess the robustness of the three-band NDVI dataset with phenological characteristics in sugarcane filed mapping,this study employed five classifiers based on four machine learning algorithms for comparative study,including Support Vector Machine classifiers based on polynomial kernel function(Polynomial-SVM),Support Vector Machine classifier based on radial basis function kernel function(RBF-SVM),Random Forest classifier(RF),Artificial Neural Network classifier(ANN)and Decision Tree classifier based on CART algorithm(CARTDT).The cloud detection results in this paper indicate that compared with other cloud-related bands/products,the Precision,Recall,Accuracy and F1-score of B1 band cloud detection results are not lower than 0.92,showing the best cloud noise detection potential.B1 band is more sensitive to cloud noise,and the detection result has the highest degree of fit with the original image,which is helpful for the design and improvement of the cloud removal algorithm of Sentinel-2 satellite image.The cloud removal results in this paper show that the cloud removal algorithm based on the Sentinel-2 image data with high temporal resolution and its aerosol band can obtain good cloud removal effect,which may be helpful to provide better image data for the study of land use/land cover change.This paper sugarcane filed recognition results showed that except for ANN classifier,Polynomial-SVM,RBF-SVM,RF,and CART-DT classifiers of sugarcane filed recognition results displayed high mapping accuracy,the overall accuracy were greater than 91.10%.The ANN classifier tended to overestimate area of sugarcane filed and underestimate area of forests.Overall performances of five classifiers suggest Polynomial-SVM has the best potential to improve sugarcane filed mapping at the regional scale.Also,this study observed that most sugarcane(more than75% of entire study area)tends to grow in flat regions with slope of less than 10°.This study discussed the unique sensitivity of aerosol band to cloud noise,verified the effectiveness of the Sentinel-2 image aerosol band cloud removal algorithm,emphasized the importance of considering phenology in rapid sugarcane filed mapping.This paper suggests that remote sensing mapping method can be used to provide reference and supplement for traditional statistical investigation,and confirms the good potential of fine-resolution Sentinel-2 images and machine learning approaches in high-accuracy land use management and decision-making.