InstaCropNet:An efficient Unet-Based architecture for precise crop row detection in agricultural applications

Autonomous navigation in farmlands is one of the key technologies for achieving autonomous management in maize fields.Among various navigation techniques,visual navigation using widely available RGB images is a cost-effective solution.However,current mainstream methods for maize crop row detection often rely on highly specialized,manually devised heuristic rules,limiting the scalability of these methods.To simplify the solution and enhance its universality,we propose an innovative crop row annotation strategy.This strategy,by simulating the strip-like structure of the crop row's central area,effectively avoids interference from lateral growth of crop leaves.Based on this,we developed a deep learning network with a dual-branch architecture,InstaCropNet,which achieves end-to-end segmentation of crop row instances.Subsequently,through the row anchor segmen-tation technique,we accurately locate the positions of different crop row instances and perform line fitting.Experimental results demonstrate that our method has an average angular deviation of no more than 2°,and the accuracy of crop row detection reaches 96.5%.

Accurate crop row recognition of maize at the seedling stage using lightweight network

Accurate extraction of crop row is very important for automation of agricultural production.Crop rows are required for accurate machine guidance in agricultural production such as fertilization,plant protection,weeding and harvesting.In this study,an efficient crop row detection algorithm called Crop-BiSeNet V2 was proposed,which combined BiSeNet V2 with a spatial convolutional neural network.The proposed Crop-BiSeNet V2 detected crop rows in color images without the use of threshold and other pre-information such as number of rows.A data set had 2697 maize crop images was constructed in challenging field trial conditions such as variable light,shadows,presence of weeds,and irregular crop shape.The proposed system was experimentally determined to overcome the interference of different complex scenes.And it can be applied to crop rows of different numbers,straight lines and curves.Different analyses were performed to check the robustness of the algorithm.Comparing this algorithm with the Fully Convolutional Networks(FCN)algorithm,it exhibited superior performance and saved 84.85 ms.The accuracy rate reached 0.9811,and the detection speed reached 65.54 ms/frame.The Crop-BiSeNet V2 algorithm proposed in this study show strong generalization performance for seedling crop row recognition.It provides high-reliability technical support for crop row detection research and assists in the study of intelligent field operation machinery navigation.

Autonomous detection of crop rows based on adaptive multi-ROI in maize fields

Crop rows detection in maize fields remains a challenging problem due to variation in illumination and weeds interference under field conditions.This study proposed an algorithm for detecting crop rows based on adaptive multi-region of interest(multi-ROI).First,the image was segmented into crop and soil and divided into several horizontally labeled strips.Feature points were located in the first image strip and initial ROI was determined.Then,the ROI window was shifted upward.For the next image strip,the operations for the previous strip were repeated until multiple ROIs were obtained.Finally,the least square method was carried out to extract navigation lines and detection lines in multi-ROI.The detection accuracy of the method was 95.3%.The average computation time was 240.8 ms.The results suggest that the proposed method has generally favorable performance and can meet the real-time and accuracy requirements for field navigation.

Automatic Guidance of an Agricultural Tractor along with the Side Shift Control of the Attached Row Crop Cultivator

This paper presents the automatic guidance system of an agricultural tractor and the side shift control of the attached row crop cultivator using electro-hydraulic actuators. In order to simulate the dynamic behaviour of the tractor along with the attached cultivator, the modified bicycle model was adopted. Steering angle sensor, fibre optic gyroscope （FOG） and RTK-DGPS technologies are assumed for measurements of the steering angle, yaw rate and the lateral position of the tractor, respectively. The kinematics model was used for the implement. In this study four cascade controllers were designed and simulated for tractor guidance which consists ofPD, PD, P and PID controllers. Other PI and PID controllers also had been designed for implement side shifting purpose. Then, these two systems were combined and the performance of the whole system was evaluated through the simulation results. According to the results tractor reaches the desired path after less than 10 seconds. Simulations showed that the maximum deviation of the tractor from the desired path was about 5 cm within this period. And the cultivator blades would follow the predetermined path with steady state error of about 5 cm too.

The Effect of Vineyard Cover Crop on Main Monomeric Phenols of Grape Berry and Wine in Vitis viniferal L. cv. Cabernet Sauvignon

This study was conducted to determine the effect of cover crop inter-row in vineyard on main mono-phenol content of grape berry and wine. Three such cover crops, two perennial legumes （white clover and alfalfa） and a perennial gramineous grass （tall fescue） were sown in vineyard. The main phenolic compounds of mature grape berry and wines vinified under the same conditions were extracted with ethyl acetate and diethyl ether and analyzed by high- performance liquid chromatography （HPLC） by comparing to soil tillage. A total of ten phenolic compounds were identified and quantified in the different grape berry and wines, including nonflavonoids （hydroxybenzoic and hydroxycinnamic acids） and flavonoids （flavanols and flavonols）. The concentration of flavonoid compounds （409.43 to 538.63 mg kg^-1 and 56.16 to 81.30 mg L^-1） was higher than nonflavonoids （76.91 to 98.85 mg kg^-1 and 30.65 to 41.22 mg L^-1） for Cabernet Sauvignon grape and wine under different treatments, respectively. In the flavonoid phenolics, Catechin was the most abundant in the different grapes and wines, accounting for 74.94 to 79.70% and 48.60 to 50.62% of total nonanthocyanin phenolics quantified, respectively. Compared to soil tillage, the sward treatments showed a higher content of main mono-phenol and total nonanthocyanin phenolics in grapes and wines. There were significant differences between two cover crop treatments （tall fescue and white clover） and soil tillage for the content of benzoic acid, salicylic acid, caffeic acid, catechin, and total phenolics in the grape berry （P 〈 0.05 or P〈0.01）. The wine from tall fescue cover crop had the highest gallic acid, caffeic acid and catechin. Cover crop system increased the total nonanthocyanin phenolics of grapes and wines in order of the four treatments： tall fescue, white clover, alfalfa, and soil tillage （control）. Cover crop in vineyard increased total phenols of grape berry and wine, and thus improved the quality of wine evidently.

Effects of Tillage and Planting Methods on Narrow and Wide Row Cotton Production

Cotton (Gossypium hirsutum L.) is an economically important crop for the Southern United States. The southern US also has a long growing season suitable for double cropping a second crop after small grains;however, the harvest date for the small grains typically occurs after the optimum planting window for cotton which reduces yield potential. A relay intercropping system was developed at Clemson University that allows interseeding of cotton into standing wheat 2 to 3 weeks before harvest with interseeded cotton yields similar to the conventional mono-cropped cotton. Therefore, the objectives of this study were 1) to determine the optimum tillage and planting methods for narrow row (76-cm) and wide row (97-cm) cotton, and 2) to compare narrow and wide row systems for conventional tillage cotton, cotton interseeded into standing wheat, and cotton planted into a terminated wheat cover crop on coastal plain soil. Two replicated tests were conducted to accomplish these objectives. In Study 1, conventional narrow row cotton combined with a deep tillage operation using Paratill yielded 23% more than conventional wide row cotton which had a deep tillage operation with a subsoiler just before planting. There were no differences between the conventional (97-cm row spacing) mono-crop and interseeded cotton yields. In Study 2, there was no significant difference in yield between narrow-row and wide-row cotton for each cropping system during the two years study. Both wide and narrow-row full season cotton had significantly higher yields than interseeded and cover crop planting systems in year two of the study. The two conservation cropping practices, wheat used as a cover crop and interseeding, showed considerable promise for reducing energy requirements, soil erosion, and wind-borne cotton damage associated with bare soil in conventional tillage. This research demonstrates the benefits of interseeding and narrow row spacing for sustainable cotton production in coastal plain soils of the Southern United States.

基于激光雷达与RGB相机融合的玉米作物行检测算法研究

针对单一传感器在面对复杂田间环境适应性差的问题,本文提出了一种基于固态激光雷达(LiDAR)与RGB相机融合的玉米作物行检测方法。首先,研究了固态激光雷达和RGB相机联合标定方法,同步获取玉米作物行图像和点云数据并进行数据预处理。然后,将预处理后的图像数据和点云数据融合,实现点云“着色”,基于点云“着色”提出聚类感兴趣密度区域算法。利用“着色”点云完成聚类,并结合作物种植农艺标准(行距),分别验证点云信息和颜色信息的可用性,能够选择最优信息完成作物行感兴趣区域聚类。最后,通过划分点云水平条带的方式确定目标点云的特征点聚类区域,取作物行特征点,并利用最小二乘法拟合作物行检测线。仅需调整行距参数,算法可实现全生命周期的作物行检测,利用正常工况下玉米苗期、前期、中期和后期数据开展算法验证,作物行中心线平均误差不大于1.781°,准确率不小于92.69%,平均耗时不超过102.7 ms。此外,为验证算法鲁棒性,开展了复杂农田背景环境,如高杂草背景、断行、苗期杂草高度与玉米高度相近以及玉米完全封行4种工况作物行检测,算法平均误差不大于1.935°,准确率不小于91.94%,平均耗时不超过108.3 ms。通过讨论阐述了基于点云“着色”开展作物行中心线提取的优越性,本文算法可为作物行中心线可靠检测提供参考。

Planting Geometry Effects on the Growth and Yield of Dryland Cotton

The declining Ogallala Aquifer beneath the Southern High Plains may necessitate dryland crop production and cotton (Gossypium hirsutum L.) is a well-adapted and potentially profitable alternative crop. The limited growing season duration of the Texas Panhandle and southwestern Kansas, however, imposes significant production risk due to incomplete boll maturation. Emphasizing earlier boll production that is usually confined to sites on lower fruiting branches may reduce risk, but offsetting high planting densities are needed to maintain desirable lint yield. Our objectives were to quantify planting: 1) row width and 2) in-row spacing effects on growth, yield, and fiber quality of dryland cotton. Field tests of row widths from 0.25 to 0.76 m and plant densities with in-row spacing ranging from 0.075 to 0.15 m were conducted from 1999 to 2005 on a nearly level Pullman clay loam (fine, mixed, superactive, thermic Torrertic Paleustoll) managed in a wheat (Triticum aestivum L.), cotton, fallow (W-Ctn-F) rotation. To expand the basis of comparison, cotton growth and yields were simulated using GOSSYM and long-term (1958-2000) weather records from Bushland, TX, as input for all combinations of 0.38 or 0.76 m row widths and plant spacing of 0.075, 0.10 and 0.15 m. Experimental and computer simulated plant height and harvested boll number increased significantly with increased row spacing and, occasionally, in-row plant spacing. Modeled lint yield for 0.38 m rows decreased by approximately 50% compared with the 582 kg·ha-1 yield for conventional row spacing, which was practically duplicated by field observations in 2001 and 2004. Measured fiber quality occasionally improved with conventional row spacing over ultra-narrow rows, but was unaffected by plant spacing. Because narrow rows and frequent plant spacing did not improve lint yield or fiber quality of dryland cotton, we do not recommend this strategy to overcome a thermally limited growing season.

基于深度学习和高斯过程回归的玉米冠下视觉导航路径提取方法

面对田间作业过程中大型机器机动性差及复杂场景下导航路径拟合精度差的问题,提出一种基于深度学习和高斯过程回归的玉米冠层下导航路径提取方法。首先,基于四足机器人采集玉米冠下作物行图像,对Mask R-CNN实例分割方法进行改进,在特征融合网络引入简化路径增强特征金字塔网络(Simple path aggregation network,Simple-PAN),通过增加自底向上的路径增强模块和特征融合操作模块,提高图像上下文特征的融合能力。其次,以模型识别的冠下作物行目标为基础构建两侧区域分界线,计算可通行区域两侧下垂叶片的分布情况,优化基于加权平均的导航路径算法。对高斯过程回归(Gaussian process regression,GPR)算法进行改进,添加DotProduct线性核对曲线拟合进行优化,优化GPR方法的直线拟合效果。最后,在验证集上进行导航路径识别,计算不同方法拟合导航路径的平均偏差。试验结果表明,该算法能够适应玉米田中叶片遮挡根茎的情况,优化的Mask R-CNN模型具备更高的冠下目标分割精度,基于改进GPR算法拟合的导航线平均偏差为0.7像素,处理一帧分辨率为1280像素×720像素的图像平均耗时为227 ms,该算法能提供在玉米冠层下具备一定避障能力的导航路径,满足导航实时性和准确性的要求。结果可为田间智能农业装备的导航算法研究提供技术与理论支撑。

基于YOLOv8和伪彩色处理的玉米去雄机导航线提取算法

抽雄期玉米的去雄打顶是提高玉米产量的重要任务,但该时期玉米的叶片、雄穗和土壤色彩难以分割,这给玉米去雄机的行间导航任务造成了困难。考虑到导航线提取的关键是在农业机械行驶区域内进行作物特征点的检测,故本文通过YOLOv8神经网络自主提取玉米雄穗感兴趣区域(ROI),并在ROI内采用Jet映射和Otsu算法分割雄穗、绿叶和土壤,然后采用FAST角点检测方法提取雄穗特征点并根据导航区域划分特征点集,最后采用最小二乘法拟合作物行检测线。实验结果表明,该算法实现了精确的、快速的玉米抽雄期导航线提取,处理单帧图像(600 pix×700 pix)平均耗时56.7 ms,平均偏差角为1.03°,能够满足玉米去雄机田间导航行驶的需求。

株间除草装置横向偏移量识别与作物行跟踪控制

株间机械除草技术与装置能有效摆脱田间除草的繁重体力劳动并消除化学除草方法所带来的危害,株间机械除草装置的牵引拖拉机在跟踪作物行时总会产生航向偏差,导致除草装置出现横向偏移,甚至无法进入除草的株间区域,同时还会增加伤苗率。为增大株间机械除草的作用区域和降低伤苗率,该文提出了通过作物行信息识别出株间机械除草装置与作物行横向偏移量的方法,并设计了株间机械除草作物行跟踪机构和控制器,实现了株间机械除草跟随作物行。采用正弦波和三角波2种标准信号作为横向偏移补偿量信号,对作物行跟踪控制器的性能进行了测试,试验结果表明:作物行跟踪控制器能较好地控制除草装置跟随横向偏移补偿信号,前进速度为0.5m/s时正弦波信号跟踪最大误差10mm,平均误差0.8mm,三角波信号跟踪最大误差11mm,平均误差1.2mm。除草试验表明,作物行跟踪控制系统能较好地控制株间除草装置跟踪作物行,在0.5m/s前进速度下跟踪最大误差为20.8mm,平均误差2.5mm;作物行跟踪控制明显减少了除草爪齿未进入株间区域的比例,在300mm株距下,可保证93.3%的株间区域有除草爪齿进行除草作业,在200mm株距下为85.9%;作物行跟踪控制降低了除草爪齿对作物的损伤,伤苗率从20%以上降到了12%以内,提高了株间机械除草的作业效果。

基于垄线平行特征的视觉导航多垄线识别

为有效快速地识别农田多条垄线以实现农业机器人视觉导航与定位,提出一种基于机器视觉的田间多垄线识别与定位方法。使用VC++6.0开发了农业机器人视觉导航定位图像处理软件。该方法通过图像预处理获得各垄行所在区域,使用垂直投影法提取出导航定位点。根据摄像机标定原理与透视变换原理,计算出各导航定位点世界坐标。然后结合垄线基本平行的特征,使用改进的基于Hough变换的农田多垄线识别算法,实现多垄线的识别与定位。使用多幅农田图像进行试验并在室内进行了模拟试验。处理一幅320×240的农田图像约耗时219.4ms,室内试验各垄线导航距与导航角的平均误差分别为2.33 mm与0.3°。结果表明,该方法能有效识别与定位农田的多条垄线,同时算法的实时性也能满足要求。

基于Census变换的双目视觉作物行识别方法

针对基于双目视觉技术的作物行识别算法在复杂农田环境下,立体匹配精度低、图像处理速度慢等问题,该文提出了一种基于Census变换的作物行识别算法。该方法运用改进的超绿-超红方法灰度化图像,以提取绿色作物行特征;采用最小核值相似算子检测作物行特征角点,以准确描述作物行轮廓信息;运用基于Census变换的立体匹配方法计算角点对应的最优视差,并根据平行双目视觉定位原理计算角点的空间坐标;根据作物行生长高度及种植规律,通过高程及宽度阈值提取有效的作物行特征点并检测作物行数量;运用主成分分析法拟合作物行中心线。采用无干扰、阴影、杂草及地头环境下的棉田视频对算法进行对比试验。试验结果表明,对于该文算法,在非地头环境下,作物行中心线的正确识别率不小于92.58%,平均偏差角度的绝对值不大于1.166°、偏差角度的标准差不大于2.628°;图像处理时间的平均值不大于0.293 s、标准差不大于0.025 s,能够满足田间导航作业的定位精度及实时性要求。

基于图像特征点粒子群聚类算法的麦田作物行检测

为了快速准确地提取麦田作物行中心线,提出了基于图像特征点粒子群聚类算法的麦田作物行检测。首先,对自然光照下获取的彩色图像运用"过绿颜色因子图像灰度化"、"Otsu图像二值化"、"左右边缘中间线检测提取作物行特征点算法"3步对图像进行预处理。然后,根据农田作物行中心线周围区域的特征点到该直线的距离均小于某一距离阈值的特征,运用粒子群优化算法对每一作物行的特征点分别进行聚类。最后,对每一类的特征点用最小二乘法进行直线拟合获取麦田作物行中心线。试验结果表明,该算法可以对作物断行、杂草、土块等复杂农田环境下的图像进行有效地作物行检测,识别率达95%,识别误差小于3°。与标准Hough算法相比,运行速率提升了一倍。该文可为实现农业机器人田间作业提供参考。

基于最大正方形的玉米作物行骨架提取算法

为了满足现代农业精准施药技术中导航路径识别的需要,该文提出一种基于最大正方形的玉米作物行骨架提取算法。首先对采集到的田间玉米作物行图像进行灰度变换,采用改进的过绿灰度化算法使作物行与背景明显分割开来;然后通过滤波、阈值分割得到二值图像;而后对经过预处理后的二值图像进行形态学中的闭运算操作,得到玉米作物行的轮廓;最后利用最大正方形准则提取玉米作物行骨架。为了验证该算法的准确度,对提取的玉米作物行骨架进行直线拟合操作,利用拟合出的中央作物行线与实际导航线偏差的大小来判断骨架提取的精准度。试验结果表明,该算法能保持骨架像素的单一性,对边缘噪声具有很强的抗干扰能力,提取骨架的误差小于5 mm,能够满足玉米对行精准施药的需求。

基于良序集和垄行结构的农机视觉导航参数提取算法

根据田间作物垄行间杂草离散的特点,基于图像矩阵,运用像素子集的良序性,结合垄宽先验知识得到垄行轨迹中心。同时,系统选择图像的绿色成分为目标特征空间,滤掉了非绿色的背景噪声,为寻找垄行子集奠定了基础。在摄像头参数结构的可线性化映射区(图像中间约1/3区域),考虑移动平台的速度和系统图像采样间隔,在系统处理速度大于平台移动速率条件下,建立了单目视觉导航系统的动态方程。试验结果表明:航向角和位置参数平均误差分别约为1°和1 mm。该算法设计简洁,实现容易,可有效避免杂草等噪声的影响,对光照也有一定的适应性。

基于机器视觉和随机方法的作物行提取算法

根据农田作物图像特点,采用基于垂直投影的窗口移动方法,得到代表作物行的定位点,提出了基于机器视觉和随机方法的作物行提取算法。该算法从定位点中随机选取2个不同点,决定一条候选直线,再根据阈值规则,进一步判断候选直线的真实性。实验结果表明,该算法可以提取不同作物的作物行,处理一幅640×480像素的彩色图像平均需要220ms,正确识别率达98%。

基于改进Hough变换的农田作物行快速检测算法

选取苗期农田作为研究对象,采集了包含行栽作物和土壤背景的农田图像,针对现有作物行定位方法易受外界干扰和处理速度较慢的不足,提出将投影法和直接Hough变换法相结合检测作物行的算法。采用2G-R-B法和OTSU法将图像二值化,通过快速中值滤波算法去除噪声,再利用垂直直方图投影将图像进行水平条划分获取作物垄平均定位点,最后通过Hough变换检测垄定位点,得到作物行中心线。试验结果表明:基于垂直直方图投影的Hough变换检测作物行中心线的算法在保证高定位精度的同时,算法处理速度比直接Hough变换检测法提高了3倍,得到的定位基准线能代表作物行走向。

基于计算机视觉的作物行定位技术

针对基于计算机视觉的作物行中心线定位困难问题,提出了基于垂直投影法的作物行定位方法。对作物图像运用过绿特征值分割作物和背景,将得到过绿特征图像划分为若干水平图像条,对图像条过绿特征值进行垂直投影,求取投影曲线上突出峰点的位置;利用稳健回归法对位置点进行线性拟合得到作物的行中心线。采用320×240像素的大豆图像进行作物行定位实验,结果表明采用该方法能够获得较好的定位结果。

基于Hough变换和Fisher准则的垄线识别算法

为了提高农业机械自主作业视觉导航的精度,基于田间作物垄行的特点,首先选择作物的绿色为特征提取垄行结构;然后针对Hough变换原理提取垄线存在的问题,根据垄线点空间关系,运用Fisher准则函数进行反压缩处理,并将Fisher准则函数值作为垄线样本点疏密程度和方向势大小的度量,优化了Hough变换识别多垄线的条件,得出了多垄识别统一模型。试验结果表明,作物垄线定位的准确性、适应性均得到提高,而且能够避免较大面积杂草等影响,从而克服了传统Hough变换提取多垄线的不足,对农田机器视觉导航应用具有一定参考价值。