苹果是我国重要的经济作物之一，准确把握苹果园种植面积，是开展果园长势监测、产量估计、农业政策制定的重要基础。然而，果园种植环境复杂，其光谱特征与林地等其他地物信息相近，传统的遥感提取方法大多都采用卫星影像数据，对遥感影像中的特征信息利用不够充分，存在提取精度不高、实时性较差等问题。本文主要结合卫星影像和无人机影像的优势，研究了一种基于“特征提取+深度学习”的苹果园快速识别提取方法，并从构建最优特征、提高模型提取精度和苹果园种植面积提取验证三个方面展开研究，具体研究内容如下：

（1）针对遥感影像中特征信息繁多，无关特征可能引入干扰信号，导致模型计算复杂度上升、泛化性能下降的问题，本文采用了基于Pearson相关系数-随机森林的双阶段特征优选策略，对Sentinel-2影像上构建的特征信息进行优选分析，以确定苹果园的最优特征组合。实验结果表明，本文特征优选策略能够有效剔除冗余信息，得到与苹果园相关性高的最优特征子集，降低了无关特征对模型提取精度的影响，为后续深度学习模型训练提供特征数据支撑。

（2）针对经典语义分割模型提取效率低，对散小目标和边缘区域易出现错分、漏分的问题，本文选取DeeplabV3+作为基础网络，对网络结构和损失函数进行改进和优化，并结合无人机数据优势，采用迁移学习的方法，构建了一种改进的DeeplabV3+语义分割模型。该模型采用轻量级MobileNetV4替代原始的特征提取网络，以提高模型提取效率；添加空间-通道注意力机制和条纹池化模块来融合局部细节和全局特征，弥补轻量级网络带来的精度损失；利用Dice Loss与Focal Loss联合损失函数缓解数据集样本类别不均衡问题；通过复用无人机影像精细化标注知识的迁移学习策略，以提高模型在Sentinel-2影像上的泛化能力。实验结果表明，本文改进的DeeplabV3+模型能够有效提高苹果园语义分割的精度和效率。

（3）为验证优选特征与改进的DeeplabV3+模型相结合对苹果园种植区面积提取的有效性，本文对苹果园种植面积提取进行了实例验证。分别对比了不同特征组合方案和不同语义分割模型对苹果园种植面积的提取精度，以及无人机影像样本种植面积与本文模型提取的苹果园种植区面积。实验结果表明，结合优选特征的改进DeeplabV3+模型对苹果园种植区分割效果相较U-Net、PSPNet、DeeplabV3+模型，错分、漏分现象更少，分割边缘更细致，面积提取的相对误差为3.36%，具有较高的提取精度，验证了本文方法的有效性。

Apple is one of the important economic crops in China, and accurately grasping the planted area of apple orchards is an important basis for orchard growth monitoring, yield estimation, and agricultural policy formulation. However, the orchard planting environment is complex, and its spectral characteristics are similar to those of other feature information such as woodland, etc. Most of the traditional remote sensing extraction methods use satellite image data, which do not fully utilize the feature information in the remote sensing images, and there are problems such as low extraction accuracy and poor real-time performance. This paper mainly combines the advantages of satellite images and UAV images, constructs a fast identification and extraction method of apple orchard based on “feature extraction + deep learning”, and carries out research in three aspects of constructing the optimal features, improving the extraction accuracy of the model, and extracting and verifying the planted area of apple orchard, with the following specific research contents:

(1) Aiming at the problem that there is a lot of feature information and irrelevant features may introduce interference signals, which leads to an increase in the computational complexity of the model and a decrease in the generalization performance, this paper adopts a two-stage feature optimization strategy based on Pearson correlation coefficient and random forest, and conducts an optimization analysis on the feature information constructed on the Sentinel-2 image in order to determine the optimal combination of features for the apple orchard. The experimental results show that the feature optimization strategy in this paper can effectively eliminate redundant information, obtain the optimal subset of features with high relevance to the apple orchard, reduce the impact of irrelevant features on the extraction accuracy of the model, and provide feature data support for the subsequent deep learning model training.

(2) Aiming at the low extraction efficiency of the classical semantic segmentation model, and the problem of easy to misclassify and omit for scattered small targets and edge regions, this paper selects DeeplabV3+ as the base network, improves and optimizes the network structure and loss function, and combines with the advantages of the UAV data, and adopts the method of migration learning to construct an improved DeeplabV3+ semantic segmentation model. The model adopts lightweight MobileNetV4 to replace the original feature extraction network to improve the model extraction efficiency; adds a spatial-channel attention mechanism and a stripe pooling module to fuse local details and global features to make up for the loss of accuracy brought by the lightweight network; and utilizes the joint loss function of Dice Loss and Focal Loss to alleviate the problem of sample category imbalance in the dataset; The migration learning strategy by reusing the fine-grained annotation knowledge of UAV images in order to improve the generalization ability of the model on Sentinel-2 images. The experimental results show that the improved DeeplabV3+ model in this paper can effectively improve the accuracy and efficiency of semantic segmentation of apple orchards.

(3) In order to verify the effectiveness of the combination of the preferred features and the improved DeeplabV3+ model for the extraction of apple orchard plantation area, this paper carries out an example validation of apple orchard plantation area extraction. The extraction accuracy of apple orchard planted area by different feature combination schemes and different semantic segmentation models, as well as the planted area of UAV image samples and the planted area of apple orchard extracted by this paper's model are compared respectively. The experimental results show that the improved DeeplabV3+ model combining the preferred features has less misclassification and omission of apple orchard plantation area segmentation compared with U-Net, PSPNet and DeeplabV3+ model, the segmentation edges are more detailed, and the relative error of area extraction is 3.36%, which is of high extraction accuracy and verifies the validity of this paper's method.

[1] 张强强, 霍学喜, 刘军弟, 等. 世界苹果产销格局及市场动态预测分析[J]. 世界农业, 2016, (07): 147-152+248.

[2] 束锡红, 樊美杉. 巩固脱贫攻坚成果同乡村振兴有效衔接案例分析——基于洛川县苹果产业发展的思考[J]. 延安大学学报(社会科学版), 2024, 46(01): 86-93+129.

[3] 张起明, 徐晗泽宇, 江丰, 等. 赣南地区柑橘果园空间分布特征分析[J]. 江西科学, 2018, 36(04): 591-598.

[4] 张继贤, 顾海燕, 杨懿, 等. 高分辨率遥感影像智能解译研究进展与趋势[J]. 遥感学报, 2021, 25(11): 2198-2210.

[5] 赵子娟, 刘东, 杭中桥. 作物遥感识别方法研究现状及展望[J]. 江苏农业科学, 2019, 47(16): 45-51.

[6] 陈仲新, 任建强, 唐华俊, 等. 农业遥感研究应用进展与展望[J]. 遥感学报, 2016, 20(05): 748-767.

[7] 戴建国, 张国顺, 郭鹏, 等. 基于无人机遥感可见光影像的北疆主要农作物分类方法[J]. 农业工程学报, 2018, 34(18): 122-129.

[8] 汪传建, 赵庆展, 马永建, 等. 基于卷积神经网络的无人机遥感农作物分类[J]. 农业机械学报, 2019, 50(11): 161-168.

[9] Wang D, Wan B, Liu J, et al. Estimating aboveground biomass of the mangrove forests on northeast Hainan Island in China using an upscaling method from field plots, UAV-LiDAR data and Sentinel-2 imagery[J]. International Journal of Applied Earth Observation and Geoinformation, 2020, 85: 101986.

[10] 邓刘洋, 沈占锋, 柯映明, 等. 基于地块尺度多时相遥感影像的冬小麦种植面积提取[J]. 农业工程学报, 2018, 34(21): 157-164.

[11] 杨颖频, 吴志峰, 骆剑承, 等. 时空协同的地块尺度作物分布遥感提取[J]. 农业工程学报, 2021, 37(07): 166-174.

[12] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521 (7553): 436-444．

[13] Zhong L, Hu L, Zhou H. Deep learning based multi-temporal crop classification[J]. Remote sensing of environment, 2019, 221: 430-443.

[14] Narkhede P R, Gokhale A V. Color image segmentation using edge detection and seeded region growing approach for CIELab and HSV color spaces[C]//2015 International Conference on Industrial Instrumentation and Control (ICIC), Pune, India, 2015: 1214-1218.

[15] Ampatzidis Y, Partel V. UAV-based high throughput Phenotyping in citrus utilizing multispectral imaging and artificial intelligence[J]. Remote Sensing, 2019, 11(04): 410.

[16] 董秀春, 刘忠友, 蒋怡, 等. 基于WorldView-2影像和语义分割模型的小麦分类提取[J]. 遥感技术与应用, 2022, 37(03): 564-570.

[17] 吕伟, 宋轩, 杨欢. 基于深度学习和多源遥感数据的玉米种植面积提取[J]. 江苏农业科学, 2023, 51(23): 171-178.

[18] 王春晓, 邢增招, 卢金莎, 等. 基于深度学习的热带水稻种植区域遥感智能提取方法研究——以海南省海口市为例[J]. 遥感技术与应用, 2024, 39(05): 1106-1114.

[19] 宋荣杰. 基于高分辨率遥感影像的果园与设施农业用地提取方法研究[D]. 咸阳: 西北农林科技大学, 2019.

[20] 陈航, 谭永忠, 邓欣雨, 等. 撂荒耕地信息获取方法研究进展与展望[J]. 农业工程学报, 2020, 36(23): 258-268.

[21] 王斌. 高精度果园空间信息自动提取方法及果园时空演变驱动因子分析[D]. 重庆: 西南大学, 2021.

[22] 高洋洋. 全色遥感影像地物信息自动分类方法研究[D]. 长春: 长春理工大学, 2017.

[23] 贾玉洁, 刘云根, 杨思林, 等. 面向Sentinel-2A影像的大理市土地利用分类方法适用性研究[J]. 浙江农林大学学报, 2022, 39(06): 1350-1358.

[24] 周珂, 杨永清, 张俨娜, 等. 光学遥感影像土地利用分类方法综述[J]. 科学技术与工程, 2021, 21(32): 13603-13613.

[25] 辛群荣, 孔维华, 胥啸宇, 等. 基于GF-1与Landsat-8影像的山区苹果园地信息提取[J]. 测绘与空间地理信息, 2017, 40(03): 104-106.

[26] 宋荣杰, 宁纪锋, 刘秀英,等.基于纹理特征和SVM的QuickBird影像苹果园提取[J]. 农业机械学报, 2017, 48(03): 188-197.

[27] Zhang J, He Y, Yuan L, et al. Machine learning-based spectral library for crop classification and status monitoring[J]. Agronomy, 2019, 9(09): 496.

[28] 顾海燕, 闫利, 李海涛, 等. 基于随机森林的地理要素面向对象自动解译方法[J]. 武汉大学学报(信息科学版), 2016, 41(02): 228-234.

[29] 吕杰, 郝宁燕, 李崇贵, 等. 利用随机森林和纹理特征的森林类型识别[J]. 遥感信息, 2017, 32(06): 109-114.

[30] 王刚. 基于landsat遥感图像的山地丘陵区果园识别与监测[D]. 昆明: 云南师范大学, 2021.

[31] Breiman L. Random forests[J]. Machine learning, 2001, 45: 5-32.

[32] 冯文卿, 眭海刚, 涂继辉, 等. 高分辨率遥感影像的随机森林变化检测方法[J]. 测绘学报, 2017, 46(11): 1880-1890.

[33] 徐晗泽宇, 刘冲, 齐述华, 等. 基于随机森林算法的赣南柑橘果园遥感信息提取[J]. 江西师范大学学报(自然科学版), 2018, 42(04): 434-440.

[34] Wang B, Li J, Jin X, et al.Mapping Tea Plantations from Multi-seasonal Landsat-8 OLI Imageries Using a Random Forest Classifier[J]. Journal of the Indian Society of Remote Sensing, 2019, 47(08): 1315-1329.

[35] 杨彦荣, 宋荣杰, 胡国强, 等. 基于随机森林和纹理特征的苹果园遥感提取[J]. 现代电子技术, 2020, 43(03): 40-44.

[36] 林娜, 王伟, 王斌. 基于随机森林和Landsat8 OLI影像的脐橙果园种植信息提取[J]. 地理空间信息, 2021, 19(11): 96-100+8-9.

[37] 赵阳. 基于遥感影像的红树林种间精细化分类方法研究与应用[D]. 天津: 国家海洋技术中心, 2023.

[38] 张东彦, 戴震, 徐新刚, 等. 基于时序Sentinel-2影像的现代农业园区作物分类研究[J]. 红外与激光工程, 2021, 50(05): 262-272.

[39] 刘羽. 基于卫星遥感的苹果果园分类和估产模型研究[D]. 西安: 西安科技大学, 2021.

[40] 牛乾坤, 刘浏, 黄冠华, 等. 基于GEE和机器学习的河套灌区复杂种植结构识别[J]. 农业工程学报, 2022, 38(06): 165-174.

[41] 韩文超, 刘明, 孙敏轩, 等. 基于标准地表光谱端元空间的苹果园种植时间制图方法[J]. 农业工程学报, 2022, 38(14): 201-210.

[42] 高若楠, 施佳子, 樊红, 等. 基于随机森林的三江平原地区作物分类研究[J]. 武汉大学学报(工学版), 2024, 57(04): 519-527.

[43] Li Z, Li H, Liu Y. A remote sensing extraction method for the area of a multi-featured citrus field in Jiangxi, Southeast China[J]. Arabian Journal of Geosciences, 2022, 15(05): 378.

[44] 陈俊杰. 基于高分辨率遥感影像纹理特征的柑橘果园提取研究[D]. 赣州: 江西理工大学, 2023.

[45] Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[C]// Computer Vision–ECCV 2014: 13th European Conference on Computer Vision, Zurich, Switzerland, 2014: 818-833.

[46] Hu W, Huang Y, Wei L, et al. Deep convolutional neural networks for hyperspectral image classification[J]. Journal of Sensors, 2015, 2015(01): 258619.

[47] Bischof H, Scneider W, Pinz AJ. Multispectural classification of Landsat-images using neual network[J], Neurocomputing, 2017, 234(10): 11-26

[48] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(06): 84-90.

[49] Penatti O A B, Nogueira K, Dos Santos J A. Do deep features generalize from everyday objects to remote sensing and aerial scenes domains?[C]//Proceedings of the IEEE conference on computer vision and pattern recognition workshops, Boston, USA, 2015: 44-51.

[50] Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 39(4): 640-651.

[51] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arxiv preprint arxiv: 2014, 1409-1556.

[52] 田萱, 王亮, 丁琪. 基于深度学习的图像语义分割方法综述[J]. 软件学报, 2019, 30(02): 440-468.

[53] Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation[C]//Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, 2015: 234-241.

[54] 孙晓敏, 郑利娟, 吴军, 等. 基于U-net的“高分五号”卫星高光谱图像土地类型分类[J]. 航天返回与遥感, 2019, 40(06): 99-106.

[55] 夏微. 基于Sentinel-2遥感数据和深度学习方法的油菜种植面积提取研究[D]. 上海: 华东师范大学, 2023.

[56] 建瑞博, 蔡智勇, 杨自尚, 等. 基于U-Net的田间小麦收获边界图像分割方法研究[J]. 河南农业大学学报, 2023, 57(03): 444-450.

[57] Chen L C, Papandreou G, Kokkinos I, et al. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 40(04): 834-848.

[58] He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(09): 1904-1916.

[59] 杨蜀秦, 宋志双, 尹瀚平, 等. 基于深度语义分割的无人机多光谱遥感作物分类方法[J]. 农业机械学报, 2021, 52(03): 185-192.

[60] 付必环. 基于深度语义分割的无人机影像烟草提取方法研究[D]. 昆明: 昆明理工大学, 2022.

[61] 董秀春, 蒋怡, 杨玉婷, 等. 基于语义分割模型和遥感的柑橘园空间信息提取[J]. 中国农机化学报, 2023, 44(01): 178-184.

[62] 常晗, 郭树欣, 张海洋,等.基于改进CBAM-DeepLab V3+的苹果种植面积提取[J]. 农业机械学报, 2023, 54(S2): 206-213.

[63] 赵玉刚, 刘文萍, 周焱, 等. 基于注意力机制和改进DeepLabV3+的无人机林区图像地物分割方法[J]. 南京林业大学学报(自然科学版), 2024, 48(04): 93-103.

[64] 武锦龙, 吴虹麒, 李浩, 等. 基于改进DeepLabV3+的荞麦苗期无人机遥感图像分割识别方法研究[J]. 农业机械学报, 2024, 55(05): 186-195.

[65] 陈馨, 孙玉军, 丁志丹. 基于Sentinel-1和Sentinel-2数据融合的森林林龄反演和动态监测[J]. 中南林业科技大学学报, 2024, 44(06): 19-29.

[66] 龙紫微, 汪泓, 贾煜, 等. 基于Sentinel-2影像的喀斯特地区土地利用信息提取[J]. 中国岩溶, 2024, 43(03): 672-683.

[67] 除多, 才旺顿珠, 拉旺顿珠, 等. Sentinel-2监测西藏冰崩灾害[J]. 遥感技术与应用, 2022, 37(06): 1289-1301.

[68] 尹晓宁, 刘兴禄, 董铁, 等. 苹果园不同覆盖材料对土壤与近地微域环境及树体生长发育的影响[J]. 中国生态农业学报, 2018, 26(01): 83-95.

[69] 王李娟, 孔钰如, 杨小冬, 等. 基于特征优选随机森林算法的农耕区土地利用分类[J]. 农业工程学报, 2020, 36(04): 244-250.

[70] 周伟, 李浩然, 石佩琪, 等. 三江源区毒杂草型退化草地植被光谱特征分析[J]. 地球信息科学学报, 2020, 22(08): 1735-1742.

[71] Mohamadi H, Habibi J, Abadeh M S, et al. Data mining with a simulated annealing based fuzzy classification system[J]. Pattern Recognition, 2008, 41(05): 1824-1833.

[72] 唐晓颖. 基于多时相遥感和GEE的苹果园识别与面积提取[D]. 泰安: 山东农业大学, 2023.

[73] 张宗海, 梁晓艳, 张玢岩. 基于灰度共生矩阵的遥感影像纹理分析方法[J]. 黑龙江科技信息, 2015, (30):36.

[74] Guo Y, Ren H. Remote sensing monitoring of maize and paddy rice planting area using GF-6 WFV red edge features[J]. Computers and Electronics in Agriculture, 2023, 207: 107714.

[75] 李剑风, 柳平增, 董超, 等. 基于哨兵二号影像和特征优选模型的生姜面积提取研究[J]. 中国农机化学报, 2025, 46(02): 207-216.

[76] 曾妍, 王迪. 极化SAR旱地作物分类的后向消除特征选择研究[J]. 中国农业资源与区划, 2023, 44(03): 222-232.

[77] 张密芳, 胡曼, 李明阳. 基于PALSAR全极化数据的城市森林蓄积量估测[J]. 南京林业大学学报(自然科学版), 2016, 40(06):5 6-62.

[78] 刘文雅, 岳安志, 季珏, 等. 基于DeepLabv3+语义分割模型的GF-2影像城市绿地提取[J]. 国土资源遥感, 2020, 32(02): 120-129.

[79] Zhao H, Shi J, Qi X, et al. Pyramid scene parsing network[C]//Proceedings of the IEEE conference on computer vision and pattern recognition, Honolulu, USA, 2017: 2881-2890.

[80] Chen L C, Zhu Y, Papandreou G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]//Proceedings of the European conference on computer vision (ECCV), Munich, Germany, 2018: 801-818.

[81] 张潜, 桑军, 吴伟群, 等. 基于Xception的细粒度图像分类[J]. 重庆大学学报, 2018, 41(05): 85-91.

[82] Woo S, Park J, Lee J Y, et al. Cbam: Convolutional block attention module[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 3-19.

[83] Qin D, Leichner C, Delakis M, et al. MobileNetV4: universal models for the mobile ecosystem[C]//European Conference on Computer Vision(ECCV), Paris, France, 2024: 78-96.

[84] 张宏鸣, 张国良, 朱珊娜, 等. 基于U-Net的葡萄种植区遥感识别方法[J].农业机械学报, 2022, 53(04): 173-182.

[85] Yu S, Wang X L. Remote sensing image segmentation method based on multilevel channel attention [J]. Laser and Optoelectronics Progress, 2020, 57(04): 142-151.

[86] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE international conference on computer vision. 2017(ICCV), Venice, Italy, 2980-2988.

[87] Milletari F, Navab N, Ahmadi S A. V-net: Fully convolutional neural networks for volumetric medical image segmentation[C]//2016 fourth international conference on 3D vision (3DV), Stanford, USA, 2016: 565-571.

[88] Pan S J, Yang Q. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10):1345-1359.

[89] 王旭涛, 陈换过, 陶瀚宇, 等. 基于迁移学习的老化风电机组故障预警方法[J]. 软件工程, 2024, 27(05): 68-72.

[90] Pan S J, Tsang I W, Kwok J T, et al. Domain adaptation via transfer component analysis[J]. IEEE transactions on neural networks, 2010, 22(02): 199-210.

[91] 张淦, 严海峰, 胡根生, 等. 基于深度学习语义分割和迁移学习策略的麦田倒伏面积识别方法[J]. 智慧农业(中英文), 2023, 5(03): 75-85.

[92] 陆泉, 何超, 陈静, 等. 基于两阶段迁移学习的多标签分类模型研究[J]. 数据分析与知识发现, 2021, 5(07): 91-100