spdis/Dorm-Air-Conditioner-Smart-Controller
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

修改一下模型相关

Browse Source
This commit is contained in:
spdis
2025-09-24 12:42:13 +08:00
parent 41e14ec828
commit 670b86d200
3 changed files with 92 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Dorm-Air-Conditioner-Smart-Controller.ino
View File

@@ -872,6 +872,12 @@ void handleSetSettings() {
prefs.putFloat("min_temp", minTemp);
prefs.putFloat("max_temp", maxTemp);
Serial.printf("温度设置已更新 - 最低: %.1f°C, 最高: %.1f°C\n", minTemp, maxTemp);
// 温度设置更改后立即执行一次判断
Serial.println("温度设置已更改，立即执行核心判断...");
executeJudgeLogic();
server.send(200, "application/json", "{\"success\":true}");
}

228
audio_model_esp32.h
View File

@@ -37,7 +37,7 @@ extern "C" {
// 预处理参数
#define MEL_FMIN 0.0f // Mel滤波器最低频率
#define MEL_FMAX 8000.0f // Mel滤波器最高频率
#define WINDOW_TYPE_HANN 1 // 汉宁窗
#define WINDOW_TYPE_HANN 1 // 海宁窗
#define ENERGY_THRESHOLD 0.01f // 音频活动检测阈值
#define CONFIDENCE_THRESHOLD 0.6f // 预测置信度阈值
@@ -100,6 +100,8 @@ float calculate_rms_energy(const int16_t* audio_data, int length);
void audio_model_cleanup(void);
const unsigned char* get_audio_model_data(void);
size_t get_audio_model_size(void);
const char* get_class_name_en(AudioClassType class_id);
const char* get_class_name_cn(AudioClassType class_id);
// ==================== 核心API函数 ====================
@@ -124,7 +126,7 @@ int audio_model_init(void) {
return -1;
}
// 预计算汉宁窗
// 预计算海宁窗
for (int i = 0; i < N_FFT; i++) {
g_preprocessor.window_buffer[i] = 0.5f * (1.0f - cosf(2.0f * M_PI * i / (N_FFT - 1)));
}
@@ -185,7 +187,7 @@ int audio_model_predict(const int16_t* audio_data, int audio_length, AudioPredic
strcpy(g_last_error, "无效参数");
return -1;
}
uint32_t start_time = micros();
// 添加看门狗喂狗
@@ -193,128 +195,73 @@ int audio_model_predict(const int16_t* audio_data, int audio_length, AudioPredic
yield();
#endif
// 使用栈上的小缓冲区替代大数组，减少内存使用
const int REDUCED_FEATURES = 32; // 减少特征数量
float mel_features[REDUCED_FEATURES];
// TODO: 集成TensorFlow Lite模型进行真实的音频识别
// 当前使用audio_model_data.h中的TensorFlow Lite模型数据
// 需要实现以下步骤：
// 1. 初始化TensorFlow Lite解释器
// 2. 加载模型数据 (audio_model_data)
// 3. 预处理音频数据为模型输入格式
// 4. 执行推理
// 5. 解析输出结果
// 简化的音频特征提取，避免复杂的Mel频谱图计算
if (preprocess_audio_to_mel_simple(audio_data, audio_length, mel_features, REDUCED_FEATURES) != 0) {
strcpy(g_last_error, "音频预处理失败");
return -1;
}
// 添加看门狗喂狗
// 临时实现：基于音频能量的简单分类，提供更合理的结果
#ifdef ARDUINO
yield();
Serial.println("警告：当前使用临时实现，等待TensorFlow Lite模型集成");
#endif
// 使用简化的特征分析替代复杂的TensorFlow Lite推理
// 计算基本统计特征
float mean_energy = 0.0f;
float energy_variance = 0.0f;
float max_energy = -1000.0f;
float min_energy = 1000.0f;
// 计算音频能量来做简单的分类判断
float rms_energy = calculate_rms_energy(audio_data, audio_length);
// 计算平均能量和能量范围
for (int i = 0; i < REDUCED_FEATURES; i++) {
mean_energy += mel_features[i];
if (mel_features[i] > max_energy) max_energy = mel_features[i];
if (mel_features[i] < min_energy) min_energy = mel_features[i];
// 定期喂狗
if (i % 10 == 0) {
#ifdef ARDUINO
yield();
#endif
// 添加调试信息：检查音频数据的实际值
#ifdef ARDUINO
int non_zero_count = 0;
int16_t min_val = 32767, max_val = -32768;
long long sum_abs = 0;
for (int i = 0; i < min(100, audio_length); i++) { // 检查前100个样本
if (audio_data[i] != 0) non_zero_count++;
if (audio_data[i] < min_val) min_val = audio_data[i];
if (audio_data[i] > max_val) max_val = audio_data[i];
sum_abs += abs(audio_data[i]);
}
Serial.printf("音频数据调试: 非零样本=%d/100, 最小值=%d, 最大值=%d, 平均绝对值=%lld\n",
non_zero_count, min_val, max_val, sum_abs/100);
#endif
// 基于能量水平进行简单分类
AudioClassType predicted_class;
float confidence;
if (rms_energy > 0.1f) {
// 高能量：可能是关门声或钥匙声
if (rms_energy > 0.3f) {
predicted_class = AUDIO_CLASS_DOOR_CLOSING;
confidence = 0.75f;
} else {
predicted_class = AUDIO_CLASS_KEY_JINGLING;
confidence = 0.65f;
}
}
mean_energy /= REDUCED_FEATURES;
// 计算能量方差
for (int i = 0; i < REDUCED_FEATURES; i++) {
float diff = mel_features[i] - mean_energy;
energy_variance += diff * diff;
}
energy_variance /= REDUCED_FEATURES;
// 添加看门狗喂狗
#ifdef ARDUINO
yield();
#endif
// 基于简化特征的分类逻辑
memset(result->class_probabilities, 0, sizeof(result->class_probabilities));
// 使用能量和方差进行简单分类
float energy_range = max_energy - min_energy;
// 钥匙声特征：中等能量，高方差
float key_score = 0.0f;
if (mean_energy > -5.0f && mean_energy < -2.0f && energy_variance > 2.0f) {
key_score = 0.4f;
} else if (rms_energy > 0.02f) {
// 中等能量：室内有人
predicted_class = AUDIO_CLASS_PERSON_PRESENT;
confidence = 0.70f;
} else {
// 低能量：室内无人
predicted_class = AUDIO_CLASS_PERSON_ABSENT;
confidence = 0.80f;
}
// 关门声特征：高能量，低方差
float door_score = 0.0f;
if (mean_energy > -2.0f && energy_variance < 1.0f) {
door_score = 0.5f;
}
// 设置结果
result->predicted_class = predicted_class;
result->confidence = confidence;
// 人员活动声特征：中等能量，中等方差
float person_score = 0.0f;
if (mean_energy > -6.0f && mean_energy < -1.0f && energy_variance > 0.5f && energy_variance < 3.0f) {
person_score = 0.3f;
}
// 无人声特征：低能量，低方差
float absent_score = 0.0f;
if (mean_energy < -8.0f && energy_variance < 0.5f) {
absent_score = 0.6f;
}
// 添加看门狗喂狗
#ifdef ARDUINO
yield();
#endif
// 归一化概率
float total_score = key_score + door_score + person_score + absent_score;
if (total_score < 0.1f) {
// 默认为有人状态
person_score = 0.4f;
total_score = 0.4f;
}
// 添加少量随机性模拟AI不确定性
uint32_t audio_hash = 0;
for (int i = 0; i < audio_length; i += 1000) {
audio_hash = audio_hash * 31 + (uint32_t)abs(audio_data[i]);
}
float noise_factor = (float)(audio_hash % 50) / 1000.0f; // 0-0.05的随机因子
result->class_probabilities[AUDIO_CLASS_KEY_JINGLING] = (key_score / total_score) + noise_factor;
result->class_probabilities[AUDIO_CLASS_DOOR_CLOSING] = (door_score / total_score) + noise_factor * 0.8f;
result->class_probabilities[AUDIO_CLASS_PERSON_PRESENT] = (person_score / total_score) + noise_factor * 0.6f;
result->class_probabilities[AUDIO_CLASS_PERSON_ABSENT] = (absent_score / total_score) + noise_factor * 0.4f;
// 重新归一化
float prob_sum = 0.0f;
// 设置概率分布
for (int i = 0; i < NUM_CLASSES; i++) {
prob_sum += result->class_probabilities[i];
}
if (prob_sum > 0) {
for (int i = 0; i < NUM_CLASSES; i++) {
result->class_probabilities[i] /= prob_sum;
}
}
// 找到最高概率的类别
result->confidence = 0.0f;
result->predicted_class = AUDIO_CLASS_PERSON_PRESENT;
for (int i = 0; i < NUM_CLASSES; i++) {
if (result->class_probabilities[i] > result->confidence) {
result->confidence = result->class_probabilities[i];
result->predicted_class = (AudioClassType)i;
if (i == (int)predicted_class) {
result->class_probabilities[i] = confidence;
} else {
result->class_probabilities[i] = (1.0f - confidence) / (NUM_CLASSES - 1);
}
}
@@ -329,6 +276,11 @@ int audio_model_predict(const int16_t* audio_data, int audio_length, AudioPredic
g_total_confidence += result->confidence;
}
#ifdef ARDUINO
Serial.printf("音频能量: %.4f, 预测类别: %s, 置信度: %.2f\n",
rms_energy, get_class_name_cn(predicted_class), confidence);
#endif
return 0;
}
@@ -406,22 +358,16 @@ int preprocess_audio_to_mel_simple(const int16_t* audio_data, int audio_length,
int16_t prev_sample = 0;
for (int i = start_idx; i < end_idx; i++) {
// 音频增益放大20倍，然后进行能量计算
int32_t amplified_sample = (int32_t)audio_data[i] * 20;
// 防止溢出，限制在int16_t范围内
if (amplified_sample > 32767) amplified_sample = 32767;
if (amplified_sample < -32768) amplified_sample = -32768;
float sample = (float)amplified_sample / 32768.0f;
float sample = (float)audio_data[i] / 32768.0f;
energy += sample * sample;
// 零交叉率计算 - 使用放大后的音频数据
// 零交叉率计算
if (i > start_idx &&
((amplified_sample >= 0 && prev_sample < 0) ||
(amplified_sample < 0 && prev_sample >= 0))) {
((audio_data[i] >= 0 && prev_sample < 0) ||
(audio_data[i] < 0 && prev_sample >= 0))) {
zero_crossings += 1.0f;
}
prev_sample = (int16_t)amplified_sample;
prev_sample = audio_data[i];
// 添加看门狗喂狗，防止长时间计算
#ifdef ARDUINO
@@ -479,22 +425,16 @@ int preprocess_audio_to_mel(const int16_t* audio_data, int audio_length, float*
int16_t prev_sample = 0;
for (int i = start_idx; i < end_idx; i++) {
// 音频增益放大20倍，然后进行能量计算
int32_t amplified_sample = (int32_t)audio_data[i] * 20;
// 防止溢出，限制在int16_t范围内
if (amplified_sample > 32767) amplified_sample = 32767;
if (amplified_sample < -32768) amplified_sample = -32768;
float sample = (float)amplified_sample / 32768.0f;
float sample = (float)audio_data[i] / 32768.0f;
energy += sample * sample;
// 零交叉率计算 - 使用放大后的音频数据
if (i > start_idx &&
((amplified_sample >= 0 && prev_sample < 0) ||
(amplified_sample < 0 && prev_sample >= 0))) {
zero_crossings += 1.0f;
}
prev_sample = (int16_t)amplified_sample;
// 零交叉率计算
if (i > start_idx &&
((audio_data[i] >= 0 && prev_sample < 0) ||
(audio_data[i] < 0 && prev_sample >= 0))) {
zero_crossings += 1.0f;
}
prev_sample = audio_data[i];
// 添加看门狗喂狗，防止长时间计算
#ifdef ARDUINO
@@ -670,13 +610,7 @@ float calculate_rms_energy(const int16_t* audio_data, int length) {
float sum = 0.0f;
for (int i = 0; i < length; i++) {
// 音频增益放大20倍，然后计算RMS能量
int32_t amplified_sample = (int32_t)audio_data[i] * 20;
// 防止溢出，限制在int16_t范围内
if (amplified_sample > 32767) amplified_sample = 32767;
if (amplified_sample < -32768) amplified_sample = -32768;
float sample = (float)amplified_sample / 32768.0f; // 归一化到[-1,1]
float sample = (float)audio_data[i] / 32768.0f; // 归一化到[-1,1]
sum += sample * sample;
}
return sqrtf(sum / length);

7
core.h
View File

@@ -188,10 +188,13 @@ int judge() {
// 获取用户设置的温度范围 (从Preferences读取)
Preferences prefs;
prefs.begin("DACSC", true); // 只读模式
float min_temp = prefs.getFloat("min_temp", 5.0); // 默认最低温度22°C
float max_temp = prefs.getFloat("max_temp", 28.0); // 默认最高温度26°C
float min_temp = prefs.getFloat("min_temp", 10.0); // 默认最低温度10°C
float max_temp = prefs.getFloat("max_temp", 28.0); // 默认最高温度28°C
prefs.end();
// 打印当前使用的温度设置
Serial.printf("当前温度设置 - 最低: %.1f°C, 最高: %.1f°C\n", min_temp, max_temp);
// 判断逻辑：基于节假日、音频识别、时间、温度和湿度的智能控制
// 规则1：节假日规则（优先级最高）