Forward looking sonar plays an important role in underwater long-distance target detection and tracking. However, forward-looking sonar image sequences suffer from low frame rates and unclear target features, which can lead to target loss. In addition, effective tracking requires compensating for sonar carrier rotation and handling target occlusion to prevent trajectory discontinuities. To solve the above problems, the characteristics of sonar image sequences and target features were combined to improve existing tracking algorithms in this paper. To mitigate target loss in sonar tracking of ByteTrack algorithm, combined with the features of forward-looking sonar images, an improved correlation method was proposed. In the first correlation, a similarity measurement method based on Kalman filter was proposed by combining motion features and target appearance features, which improved the accuracy of tracking. To address rapid apparent target motion caused by sonar carrier rotation, rotation compensation was added to the ByteTrack algorithm using sonar attitude data to improve the accuracy of matching; Finally, the superiority of the improved association method and the combination of target appearance features were demonstrated through comparative experiments using similarity measurement algorithms. Compared with mainstream target tracking algorithms such as DeepSort, TransTrack, and ByteTrack, the improved model achieved a tracking accuracy of 76.8% and a tracking recall rate of 80.6%. Compared with the original ByteTrack algorithm, the improved ByteTrack has improved tracking accuracy by 9.4%, recall by 10.8%, and ID switching frequency by 46%. The fusion experimental results of detection and tracking show that the improved target detection tracking fusion algorithm has lower miss rate, lower false alarm rate, lower identity switching times, making it well-suited for underwater target detection and tracking with forward-looking sonar.