基于RetinaNet模型的鸟巢智能检测

为解决架空输电线路运行过程中因设备及杆塔上鸟巢对输电线路造成的不良影响,本文通过对比分析一阶目标检测模型和二阶目标检测模型的优劣,选取以分类损失函数为核心、特征金字塔网络为骨干网络的RetinaNet模型用于鸟巢目标的自动检测。解决了经典的一阶目标检测模型和二阶目标检测模型对鸟巢的检测效率或准确率比较低的问题。本文实验首先通过数据集选取及数据集预处理,并经过模型训练逐步优化调整网络结构和参数,建立了适合鸟巢检测的RetinaNet模型,实现对鸟巢的快速准确检测。实验结果表明,RetinaNet模型对输电线路的鸟巢的的平均准确率为94.1%,每张图片的识别速度为68ms,通过与Faster R-CNN、YOLO及SSD方法进行比较,验证了RetinaNet模型对输电线路设备及杆塔上鸟巢检测的有效性和可靠性。

Intelligent detection of bird

In order to solve the adverse effect of the bird''s nest on the transmission line caused by the equipment and tower during the operation of the overhead transmission line, this paper compares and analyzes the advantages and disadvantages of the first-order target detection model and the second-order target detection model, and selects the RetinaNet model with the classification loss function as the core and the characteristic pyramid network as the backbone network for the automatic detection of the bird''s nest target. It solves the problem that the classical first-order target detection model and the second-order target detection model have low detection efficiency or accuracy. In this experiment, firstly, through data set selection and data set preprocessing, and through model training, the network structure and parameters are optimized and adjusted step by step, and a RetaiNet model suitable for nest detection is established to achieve rapid and accurate nest detection. The experimental results show that the average accuracy of RetinaNet model to the bird''s nest of transmission line is 94.1%, and the recognition speed of each picture is 68ms. By comparing with Faster R-CNN, YOlO and SSD methods, the validity and reliability of RetinaNet model to the bird''s nest detection on transmission line equipment and tower are verified.