✨ 长期致力于树木年龄、微损、针刺仪、抗钻阻力值序列研究工作，擅长数据搜集与处理、建模仿真、程序编写、仿真设计。
✅ 专业定制毕设、代码
✅如需沟通交流，点击《获取方式》
（1）平稳卡尔曼滤波器去噪与年龄估计：
根据卡尔曼滤波理论，建立针刺仪抗钻阻力值序列的状态空间模型。状态变量为真实阻力值和其一阶导数，观测值为含噪声测量值。推导平稳卡尔曼滤波器，稳态增益仅依赖于参数Rat = Q/R，其中Q为过程噪声方差，R为测量噪声方差。Rat取值范围0.05至0.5，根据胸径大小选择：胸径小于20cm取0.05，20-40cm取0.2，大于40cm取0.35。滤波后序列的波峰波谷计数即为年龄。对华北落叶松323组数据应用，原始自动判读平均相对误差-40.49%，本算法降至0.75%，t检验t=-0.4682表明无显著差异。
（2）频谱分析算法确定年轮周期：
对抗钻阻力值序列去除线性趋势后，应用离散傅里叶变换。找出幅度谱中最大幅值对应的谐波阶次，其周期数作为年龄估计。引入窗口参数Wid控制去趋势窗口长度，Wid取101、201或301。对华北落叶松，Wid=201效果最佳。平均相对误差-0.35%，t值0.85。分析表明，胸径40cm以上建议Wid=301，20-40cm用201，20cm以下用101。
（3）峰谷分析结合BP神经网络预测阈值：
峰谷分析算法通过检测局部极大值和极小值计数，需要阈值参数Det过滤无效波动。Det取值范围0.02-0.15。利用BP神经网络自动预测Det，输入特征包括10个统计量：均值、标准差、偏度、峰度、最大一阶差分、零交叉率、自相关系数等。网络结构10-6-1，训练集80组，测试集33组。预测Det与实际最优Det的绝对误差平均0.008。应用该算法于红松、臭冷杉、长白落叶松113组数据，平均相对误差-2.69%，t值1.31。
import numpy as np from scipy import signal, fft import tensorflow as tf from sklearn.preprocessing import StandardScaler class StationaryKalman: def __init__(self, Rat=0.2): self.Rat = Rat # Q/R ratio # 稳态增益计算: 对于二阶系统，已知解析解 self.K = self._compute_gain() def _compute_gain(self): # 简化: 稳态增益矩阵 [K1, K2] qr = self.Rat K1 = (np.sqrt(1+4*qr) - 1) / (2*qr) K2 = (np.sqrt(1+4*qr) - 1) / (2) return np.array([K1, K2]) def filter(self, z): n = len(z) x_est = np.zeros(n) x_est[0] = z[0] prev_x = np.array([z[0], 0]) for k in range(1, n): # 预测 x_pred = np.array([prev_x[0] + prev_x[1], prev_x[1]]) # 更新 y_res = z[k] - x_pred[0] x_new = x_pred + self.K * y_res x_est[k] = x_new[0] prev_x = x_new return x_est def count_peaks_valleys(self, x): peaks = signal.find_peaks(x)[0] valleys = signal.find_peaks(-x)[0] return len(peaks) + len(valleys) class SpectralAgeEstimator: def __init__(self, window_len=201): self.Wid = window_len def detrend(self, series): from scipy import signal b = np.ones(self.Wid)/self.Wid trend = signal.convolve(series, b, mode='same') return series - trend def estimate_age(self, series): detrended = self.detrend(series) f = fft.fft(detrended) magnitude = np.abs(f[:len(series)//2]) # 找到最大幅值对应的频率索引，排除0频 peak_idx = np.argmax(magnitude[1:]) + 1 period = len(series) / peak_idx return int(round(period)) class BP_Predictor: def __init__(self): self.model = tf.keras.Sequential([ tf.keras.layers.Dense(6, activation='tanh', input_shape=(10,)), tf.keras.layers.Dense(1, activation='sigmoid') ]) self.model.compile(optimizer='adam', loss='mse') self.scaler = StandardScaler() def extract_features(self, series): features = [] features.append(np.mean(series)) features.append(np.std(series)) features.append(3*(np.mean(series)-np.median(series))/np.std(series)) # 偏度近似 features.append(np.mean((series-np.mean(series))**4)/(np.std(series)**4)-3) # 峰度近似 features.append(np.max(np.diff(series))) zero_cross = np.sum(np.diff(np.sign(series)) != 0) features.append(zero_cross / len(series)) acf = np.correlate(series, series, mode='full') acf = acf[len(series)-1:] / acf[len(series)-1] features.append(acf[1] if len(acf)>1 else 0) features.append(acf[2] if len(acf)>2 else 0) features.append(acf[3] if len(acf)>3 else 0) features.append(acf[4] if len(acf)>4 else 0) return np.array(features) def train(self, X, y): X_scaled = self.scaler.fit_transform(X) self.model.fit(X_scaled, y, epochs=200, verbose=0) def predict(self, series): feat = self.extract_features(series) feat_scaled = self.scaler.transform(feat.reshape(1,-1)) det = self.model.predict(feat_scaled, verbose=0)[0,0] return det * 0.15 # 映射到0-0.15范围 class PeakValleyAnalyzer: def __init__(self, det=0.05): self.det = det def count_age(self, series): # 局部极值检测，阈值Det diff = np.diff(series) sign_change = np.where(np.diff(np.sign(diff)) != 0)[0] # 过滤振幅小于det的波动 valid = [] for idx in sign_change: if idx+1 < len(series): if abs(series[idx+1] - series[idx]) > self.det: valid.append(idx) return len(valid)
