Python基于隐马尔科夫模型的笔势识别源码(课程设计).zip

from data import train_set, val_set, gesture_list from kmeans import clusters, K import copy import math from hmmlearn import hmm import numpy as np # def get_cluster_num(point): # x = point[0] # y = point[1] # for cluster_num in range(len(clusters)): # for cluster_point in clusters[cluster_num]: # if cluster_point[0] == x and cluster_point[1] == y: # return cluster_num # return K def get_cluster_num(point): # 用k=1的KNN得到测试集每个点在Kmeans上的观测值 x = point[0] y = point[1] distances_cluster_nums = [] for clsuter_num in range(len(clusters)): for cluster_point in clusters[clsuter_num]: distance = math.sqrt( (cluster_point[0] - x) ** 2 + (cluster_point[1] - y) ** 2 ) distances_cluster_nums.append((distance, clsuter_num)) min_distance_cluster_num = min(distances_cluster_nums, key=lambda dis: dis[0]) return min_distance_cluster_num[1] train_obs = { 'a': [], 'e': [], 'i': [], 'o': [], 'u': [], } val_obs = { 'a': [], 'e': [], 'i': [], 'o':[], 'u':[], } for gesture_name in gesture_list: for word in train_set[gesture_name]: word_obs = [] for word_point in word: cluster_num = get_cluster_num(word_point) word_obs.append(cluster_num) train_obs[gesture_name].append(word_obs) for gesture_name in gesture_list: for word in val_set[gesture_name]: word_obs = [] for word_point in word: cluster_num = get_cluster_num(word_point) word_obs.append(cluster_num) val_obs[gesture_name].append(word_obs) model = {} # 隐藏状态数量 n_states = 3 # 观测状态数量（Kmeans的K值） n_observations = K for gesture_name in gesture_list: model[gesture_name] = hmm.CategoricalHMM(n_components=n_states, n_iter=100) model[gesture_name].n_features = n_observations observations = train_obs[gesture_name] observations = [np.array(obs).reshape(-1, 1) for obs in observations] lengths = [obs.shape[0] for obs in observations] # print(lengths) model[gesture_name].fit(np.concatenate(observations), lengths) val_obs = { 'a': [], 'e': [], 'i': [], 'o':[], 'u':[], } for gesture_name in gesture_list: for word in val_set[gesture_name]: word_obs = [] for word_point in word: cluster_num = get_cluster_num(word_point) word_obs.append(cluster_num) val_obs[gesture_name].append(word_obs) stat = { 'a': [], 'e': [], 'i': [], 'o':[], 'u':[], } for gesture_name in gesture_list: for obs in val_obs[gesture_name]: obs_seq = np.array(obs).reshape(-1, 1) log_probs = [] for model_gesture in model: log_prob = model[model_gesture].score(obs_seq) log_probs.append((log_prob, model_gesture)) # 选出概率最大的模型 max_prob_model_gesture = max(log_probs, key=lambda prob_tuple: prob_tuple[0])[1] stat[gesture_name].append(max_prob_model_gesture) confusion_stat = {} record_lengths = {'a': 0, 'e': 0, 'i': 0, 'o': 0, 'u': 0,} # 对于每一个字母测试用例的数量 for stat_key in stat: count = {'a': 0, 'e': 0, 'i': 0, 'o': 0, 'u': 0,} for record in stat[stat_key]: count[record] += 1 record_lengths[stat_key] += 1 confusion_stat[stat_key] = count confusion_matrix = copy.deepcopy(confusion_stat) for vowel_true in confusion_matrix: for vowel_infer in confusion_matrix[vowel_true]: confusion_matrix[vowel_true][vowel_infer] /= record_lengths[vowel_true] # 画混淆矩阵 print(f'聚类数量为{K}，隐藏状态数量为{n_states}') print('\ta\te\ti\to\tu') for vowel_true in gesture_list: print(vowel_true + '\t', end='') for vowel_infer in gesture_list: print(confusion_matrix[vowel_true][vowel_infer], '\t', end='') print() if __name__ == '__main__': import matplotlib.pyplot as plt # colors = ['b', 'g', 'r', 'c', 'm'] # plt.figure(figsize=(4, 4)) # for k in range(5): # for point in clusters[k]: # plt.scatter(point[0], point[1], c=colors[k]) # plt.show()