Aiming at the limitations of the sample weight updating mechanism of the traditional Adaboost algorithm, the article proposes three improved algorithms based on the joint weight updating mechanism to solve the problems of sample distribution imbalance, etc. The MW_UA algorithm is centered on the updating of the proportion of the sample weight, the OW_UA algorithm realizes the updating of the weight of the sample set based on the classification effect of the initial samples, the MAR_UA algorithm employs sample The MAR_UA algorithm uses the sample Margin to quantify the degree of difficulty of sample classification and then obtain the corresponding sample weights. The performance test experiments and prediction simulation experiments of the improved algorithm are based on the MWSP and Caltech datasets. The experimental results show that the average accuracy and F1 score of MAR_UA algorithm in the two datasets are over 90%, which is the best performance among all the improved algorithms. The algorithm also shows optimal prediction error convergence performance in both datasets, and the training error can be converged to the minimum within 40 times of training. When the algorithm is applied to the simulation experiment of pedestrian recognition, it has the best recognition effect in the sunny environment, with a detection rate of 94.1%. In addition, the error between its predicted and real values of offshore wind speed is no more than 0.2 m/s, and the ERMS and EMA are reduced by 63.52% and 55.5%, respectively, compared with the traditional Adaboost. This study optimizes the weight updating mechanism of the joint Adaboost algorithm using various methods, which can provide new ideas for the optimization research of the weight updating mechanism.