基于深度学习的二值图像目标轮廓识别算法

针对传统目标轮廓识别算法对图像目标轮廓识别精度较低、 效果较差的问题, 提出一种基于深度学习的二值图像目标轮廓识别算法. 首先, 选取深度学习算法中的深度卷积网络算法识别二值图像目标轮廓, 将二值图像划分为不重叠的、 大小相同的子块图像输入深度卷积网络第一层； 其次, 卷积网络中的滤波器（卷积核）采用传统神经网络算法优化的代价函数对输入子块图像实施卷积滤波, 并将卷积滤波后下采样图像发送至第二层, 第二层经过相同处理后将结果输入第三层, 第三层输出图像即为该子块目标轮廓识别结果； 最后, 所有子块识别结束后在输出层通过全连接方法将其聚类, 并输出最终二值图像目标轮廓识别结果. 实验结果表明, 该算法识别15幅二值图像目标轮廓的识别精度平均为98.75%, 信噪比平均为2.42, 识别效果较优.

Binary Image Target Contour Recognition Algorithm Based on Deep Learning

Aiming at the problem of low accuracy and poor effect of traditional target contour recognition algorithm for image target contour recognition, the author proposed a binary image target contour recognition algorithm based on deep learning. Firstly, the deep convolution network algorithm in the deep learning algorithm was selected to recognize the target contour of binary image, and the binary image was divided into non overlapping sub block images with the same size and input into the first layer of the deep convolution network. Secondly, the filter (convolution kernel) in the convolution network used the cost function optimized by the traditional neural network algorithm to implement convolution filtering on the input sub block image, and sent the down sampling image after convolution filtering to the second layer. After the same processing, the second layer input the result into the third layer, and the output image of the third layer was the target contour recognition result of the sub block. Finally, after all the sub blocks were recognized, all the sub blocks were clustered in the output layer by the full connection method, and the final binary image target contour recognition results were output. The experimental results show that the average recognition accuracy of 15 binary images is 98.75%, the average signal-to-noise ratio is 2.42, and the recognition effect is better.