The Spatial Distribution Extraction Method Of Winter Wheat Based On GF-6 WFV

Winter wheat spatial distribution information is critical for yield estimation and is the cornerstone for large-scale modification of winter wheat planting structures and policy design.We can dynamically monitor farmland moisture and crop distribution to satisfy the demands of wide-scale surface information detection by employing remote sensing satellites to gather timely agricultural photos and convolutional neural networks to extract farmland information in broad regions.In recent years,multispectral data has provided a reliable data source for obtaining highprecision spatial distribution of winter wheat with a larger number of bands,a narrower band range,and richer remote sensing information.However,the types of features covered by remote sensing images are complex,and there are phenomena such as "same thing,different spectra," and "same spectrum,different thing," and the identification of planting plot boundaries is blurred in the reanalysis.The GF-6 WFV is used as the data source in this article,and the data features are analyzed to develop a semantic segmentation model that integrates multi-scale and multi-level feature outputs.The primary research topics and outcomes are as follows:1.The spectral curve features,spectral statistical features,and spectral correlation features are studied for the GF-6 WFV data,and the significance is enhanced based on the spectral correlation features.The enhancement outcomes are evaluated.Six strategies are developed based on the study to add more divergent and separable data features to the model and provide a priori knowledge for semantic segmentation.The experimental results show that the F1 scores of the six schemes with feature vectors are 2.51%,1.28 percent,2.31 percent,1.32 percent,2.87 percent,and 1.06 percent higher than the 8-band input without feature vectors,indicating that feature vectors based on data feature analysis and combined feature vectors can improve extraction accuracy of winter wheat.2.To address the issue of ambiguous boundary identification of different categories in semantic segmentation results,the CMFNet（Converged Multi-layer Feature Network）semantic segmentation model is proposed,which fuses multi-scale and multi-layer feature output results to obtain the final output by preserving detailed information lost due to perceptual field expansion and fusing it with high-level semantic information.The experimental results demonstrate that CMFNet’s F1 score is 85.09 percent,which is higher than the comparison model in the experiment,demonstrating that using multi-layer features improves extraction accuracy.3.To eliminate the extraction accuracy error caused by misidentifying water bodies during the extraction of spatial distribution information of winter wheat,the spectra with the most obvious response to water bodies were statistically analyzed using spectral curves,and water bodies were removed using the threshold method.These findings can be used to extract crop spatial distribution information using GF-6WFV data on a wide scale,rapidly and finely,and to acquire accurate planting data.They also provide a foundation for extracting crop spatial distribution information using GF-6 WFV data.