Abstract: The traditional seeker optimization algorithm (SOA) uses three steps for an optimal search:calculating the search direction, searching the step length, and updating the individual position. Its shortcomings are the large amount of calculation required and weak communication between populations, which results in low speed optimization. To address these disadvantages, this paper offers the binomial crossover operator improved seeker optimization algorithm (BCOISOA) as an improvement. In terms of computational search step length, this paper adopts a random number and maximum function product judgment subgroup location so that global optimization computation speed can be improved. In terms of update location, this paper puts forward two crossover operators to strengthen the connection between the populations. This avoids premature convergence of the algorithm during the process of updating the search direction, caused by the local optimum, and achieves a fast and accurate optimal solution. This article usesthe BCOISOA-BP neural network algorithm for a two-phase grinding process to achieve a grind size online soft sensor. Compared with the SOA and PSO algorithms, the simulation result shows that the BCOISOA algorithm has the fastest convergence speed and highest precision. It therefore satisfies the requirements of grind size real-time detection.