农机专利成果运营库目标检测数据

数据集可用于农业机械自动化、智能农业和精准农业相关的目标检测模型训练。以农机视角为主的、烟草和棉花为代表的大叶作物的识别和定位问题，用于验证相关专利的可行性，还能为智能农业领域提供可靠的植株检测基础，为精准农业管理和自动化操作提供数据支持。本数据是评估若干专利进行专利运营时的内部测试数据脱敏后的目标检测数据集，数据采用地面手拍模仿农机和移动机器人视角，在多个农田拍摄图像，高度从地面30cm到150cm不等。使用多角度、多高度的采集方式确保了数据的多样性，涵盖了不同生长阶段、光照条件和种植密度的烟草和棉花植株。对图像进行预处理，保留原始分辨率和特征，以确保植株的自然特征不被改变。同时，应用随机亮度调整、植株对比度增强、随机旋转、随机裁剪等图像增强技术提高模型的泛化能力。使用植株对比度增强算法，通过调整图像像素值相对于图像均值的差异来实现，使植株轮廓更加清晰，有助于提高模型对植株的识别能力，尤其是在复杂背景或光照条件不佳的情况下。采用COCO格式进行双类目标检测标注，标注过程使用包围框(x,y,width,height)对每个植株进行精确定位。将COCO格式转换为CSV格式：
x1=x,y1=y,x2=x+width,y2=y+height,cls∈{Tobaccoplant,Cottonplant}，分别对应烟草植株和棉花植株。cls为类别，x1,y1,x2,y2是坐标。

This dataset is intended for training object detection models related to agricultural machinery automation, smart agriculture, and precision agriculture. It addresses the recognition and localization of broad-leaf crops, typified by tobacco and cotton, from the perspective of agricultural machinery. It can be used to verify the feasibility of relevant patents, provide a reliable plant detection foundation for the smart agriculture field, and offer data support for precise agricultural management and automated operations.

This data is a desensitized object detection dataset derived from internal test data for patent operation evaluation of several patents. Images were captured in multiple farmlands via hand-held shooting from the ground to simulate the perspectives of agricultural machinery and mobile robots, with shooting heights ranging from 30 cm to 150 cm above the ground. The multi-angle and multi-height acquisition method ensures data diversity, covering tobacco and cotton plants at different growth stages, lighting conditions, and planting densities.

Image preprocessing was conducted while retaining the original resolution and features to preserve the natural characteristics of the plants. Meanwhile, image augmentation techniques including random brightness adjustment, plant contrast enhancement, random rotation, and random cropping were employed to enhance the model's generalization capability. The plant contrast enhancement algorithm works by adjusting the disparity between image pixel values and the image mean, which sharpens plant contours and helps improve the model's plant recognition performance, especially in complex backgrounds or under suboptimal lighting conditions.

Dual-class object detection annotations were carried out in COCO format, with each plant precisely localized using bounding boxes (x, y, width, height). The COCO format annotations were then converted to CSV format with the following rules: x1 = x, y1 = y, x2 = x + width, y2 = y + height, where cls ∈ {Tobaccoplant, Cottonplant}, corresponding to tobacco plants and cotton plants respectively. Here, cls denotes the object category, and x1, y1, x2, y2 represent the bounding box coordinates.

搜集汇总

数据集介绍

特点

该数据集是一个针对农业机械自动化和智能农业的目标检测数据集，包含8065条烟草和棉花植株的图像数据，采用多角度、多高度采集方式，并经过预处理和增强处理，标注信息包括植株类别和位置，适用于相关专利验证和智能农业应用。