Weeds are plants that harm crops by inhibiting the growth of cultivated plants. The first step to take control of weeds is by identifying weed among the cultivating plant. The fastest and easiest way to identify plants, including weeds is by its leaves. This research proposing weed species recognition based on weeds leaf images by extracting its shape and texture features. Moment invariant method is used to get the shape and Lacunarity method for the texturel.? Neural Network with backpropagation learning algorithm are implements for the extracted features recognition proses. The result of this research achievement shows the highest level of recognition accuracy of 97.22% before the noise is eliminated in the image of the Canny edge detection. Highest level of accuracy is obtained using two features from moment invariant (moment ?and? ) and 1 lacunarity’s feature (size box 4 x 4 or 16 x 16). The use of 3 neurons in the hidden layer of Artificial Neural Network (ANN) provide training time data more quickly than by using 1 or 2 hidden layer neurons.
展开▼
机译:杂草是通过抑制栽培植物的生长而伤害作物的植物。控制杂草的第一步是通过在栽培植物中识别杂草。识别植物(包括杂草)的最快,最简单的方法就是叶子。这项研究提出了通过提取杂草叶片图像的形状和纹理特征来识别杂草种类的技术。矩不变法用于获得纹理的形状和腔隙度法。具有反向传播学习算法的神经网络是提取特征识别过程的工具。这项研究成果的结果表明,在消除Canny边缘检测图像中的噪声之前,最高的识别精度为97.22%。使用矩不变(矩?和?)和1个盲点的特征(大小框4 x 4或16 x 16)的两个特征可以获得最高的准确性。与使用1个或2个隐藏层神经元相比,在人工神经网络(ANN)的隐藏层中使用3个神经元可以更快地提供训练时间数据。
展开▼