番茄病虫害高效识别图像标注数据

数据可用于农作物病害检测、叶片计数等相关的目标检测和图像分类模型训练。可应用于番茄作物叶片病害的高效识别和计数，帮助种植业从业者提前做好相关的病害预防措施，保障产量；对通用的作物病害防控具有重要的数据价值。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，涵盖了不同光照条件、生长阶段和病害程度的番茄叶片图像。图像预处理，基于原始分辨率和特征，通过随机亮度调整-对比度数据增强-随机旋转-随机缩放-随机翻转来增加图像，提高模型的泛化能力，确保叶片的纹理和病害特征不被改变。对比度增强：将图像像素值减去图像均值，乘以一个随机生成的对比度因子，再加回图像均值。在保持图像整体亮度不变的情况下，增强或减弱图像的对比度。通过随机调整对比度，可以模拟不同光照条件下叶片的视觉效果，增强模型对各种实际场景的适应能力采用COCO包围框格式进行目标检测标注，标注过程使用包围框(x,y,width,height)对每个叶片进行精确定位。将COCO格式转换为CSV格式：x1=x,y1=y,x2=x+width,y2=y+height,cls∈{"Earlyblightleaf","Septorialeafspot","bacterialspot","lateblight","mosaicvirus","yellowvirus","moldleaf"}，cls为类别，x1,y1,x2,y2是坐标。

This dataset can be used for training object detection and image classification models for tasks including crop disease detection and leaf counting. It can be applied to efficient identification and counting of tomato leaf diseases, helping crop farming practitioners take targeted disease prevention measures in advance and ensure crop yield, and holds important data value for general crop disease prevention and control.

This is a desensitized object detection dataset derived from internal test data generated during the evaluation of several patents for patent operation, covering tomato leaf images under different lighting conditions, growth stages and disease severities.

For image preprocessing, based on the original resolution and features, the dataset is augmented via a series of operations including random brightness adjustment, random contrast adjustment, random rotation, random scaling and random flipping, to improve the generalization ability of the model while ensuring that leaf textures and disease-related features remain unchanged.

Contrast enhancement: Subtract the image mean value from the pixel values of the image, multiply by a randomly generated contrast factor, then add back the image mean value. This operation can enhance or reduce the image contrast while keeping the overall image brightness unchanged. By randomly adjusting the contrast, the visual effects of leaves under different lighting conditions can be simulated, enhancing the model's adaptability to various real-world scenarios.

Object detection annotations are conducted in COCO bounding box format, where each leaf is accurately positioned using bounding boxes in the form of (x, y, width, height) during the annotation process.

The COCO format is converted to CSV format as follows: x1 = x, y1 = y, x2 = x + width, y2 = y + height, where cls ∈ {"Earlyblightleaf", "Septorialeafspot", "bacterialspot", "lateblight", "mosaicvirus", "yellowvirus", "moldleaf"}, cls represents the category, and x1, y1, x2, y2 are the bounding box coordinates.