BPNN is a multi-layer network with forward and backward error algorithms. It is currently the most widely used NN. Especially when it is applied to the basketball teaching scene, it can improve the research on the parameter optimization and state evaluation of the BP neural network in the teaching scene. This paper is mainly to study how to optimize the parameters of basketball teaching scene to improve the teaching effect and to evaluate the state. This paper improves the BPNN technology to study the basketball teaching scene and then proposes a genetic algorithm on the basis of the BPNN. It optimizes the parameters of basketball teaching scene through genetic algorithm and then analyzes the experimental data. The experimental results of this paper show that the BPNN is not as good as the BPNN based on the genetic algorithm in optimizing the parameters of the basketball teaching scene. The BPNN based on genetic algorithm has a successful recognition rate of more than 93% for basketball dribbling and shooting, which is 7-8% higher than that of the BPNN.