Research On Underwater Fish Target Detection Algorithm Based On Bionic Dolphin

At present,there are two types of target detection algorithms in the field of computer vision: two-stage and one-stage.The former has excellent detection accuracy,but the detection speed is slow,the network model is large and the hardware equipment has high requirements.The comparison is suitable for video analysis scene,but not for real-time target detection.Although the accuracy of the latter is not as good as that of the former,the speed of detection can meet the real-time requirements,but there are still certain requirements for the computing power of the equipment and the model is large.However,for the specific scenario of deployment to the mobile end,the computing resources of the device are limited.One-stage and two-stage algorithms can not be directly applied to mobile terminals.Therefore,a real-time target detection algorithm,which can guarantee a certain detection accuracy and has a low demand for equipment calculation,is of great application value.At present,there are more and more researches on lightweight target detection model at home and abroad.This paper mainly studies the underwater fish target detection algorithm applied to the mobile end.According to the requirements of the planning,the detection algorithm in this paper can be applied to the bionic dolphin mobile terminal.The detection algorithm can improve the detection speed to meet the real-time requirements and reduce the power consumption of the algorithm under the premise of ensuring certain accuracy.The research work of this paper is as follows:Firstly,the model needs high-quality training data.Therefore,after collecting the original data,the data enhancement includes geometric transformation,color dithering and noise processing.In order to enhance the diversity of data,we use the generative adversarial network to transfer some data styles and replace different underwater background images.Finally,we get enough training data.Then it introduces the advanced target detection algorithm and analyzes its principle.Then,according to the comparative analysis of the advantages of each algorithm,select the reference algorithm SSD which is suitable for this scenario and on this basis change its backbone network to Shuffle Net.It makes the model meet the requirements of lightening and improving detection speed while ensuring certain detection accuracy.The,performance of the algorithm is further optimized after target detection combined with target tracking function.This paper introduces the KCF algorithm which performs well in tracking speed and progress.The advantage of target tracking is that it needs less time and computation.Using trace to reduce the number of times of detection can reduce the computing load of the processor,thus reducing the power consumption of the system.At the same time,the detection time is shortened to further guarantee the real-time performance and avoid the problem that the detection algorithm may miss detection in the next frame.Finally,the whole program is written.Including image data enhancement code,generation of anti network code,detection of neural network construction,tracking and detection code.Combined with the function of target detection and target tracking,the test results are obtained.Finally,the map of underwater fish target detection algorithm in this paper is 66.7.