In this paper, an improved Fast-R-CNN nuclear power cold source disaster biological image recognition algorithm is proposed to improve the safety operation of nuclear power plants. Firstly, the image data sets of the ...In this paper, an improved Fast-R-CNN nuclear power cold source disaster biological image recognition algorithm is proposed to improve the safety operation of nuclear power plants. Firstly, the image data sets of the disaster-causing creatures hairy shrimp and jellyfish were established. Then, in order to solve the problems of low recognition accuracy and unrecognizable small entities in disaster biometrics, Gamma correction algorithm was used to optimize the image of the data set, improve the image quality and reduce the noise interference. Transposed convolution is introduced into the convolution layer to increase the recognition accuracy of small targets. The experimental results show that the recognition rate of this algorithm is 6.75%, 7.5%, 9.8% and 9.03% higher than that of ResNet-50, MobileNetv1, GoogleNet and VGG16, respectively. The actual test results show that the accuracy of this algorithm is obviously better than other algorithms, and the recognition efficiency is higher, which basically meets the preset requirements of this paper.展开更多
文摘In this paper, an improved Fast-R-CNN nuclear power cold source disaster biological image recognition algorithm is proposed to improve the safety operation of nuclear power plants. Firstly, the image data sets of the disaster-causing creatures hairy shrimp and jellyfish were established. Then, in order to solve the problems of low recognition accuracy and unrecognizable small entities in disaster biometrics, Gamma correction algorithm was used to optimize the image of the data set, improve the image quality and reduce the noise interference. Transposed convolution is introduced into the convolution layer to increase the recognition accuracy of small targets. The experimental results show that the recognition rate of this algorithm is 6.75%, 7.5%, 9.8% and 9.03% higher than that of ResNet-50, MobileNetv1, GoogleNet and VGG16, respectively. The actual test results show that the accuracy of this algorithm is obviously better than other algorithms, and the recognition efficiency is higher, which basically meets the preset requirements of this paper.
