Abstract Acoustical detection of insects feeding and crawling sounds was used to automatically monitor internal and external grain feeding bruchids in order to assess the growth and density of food legume bruchids (Callosobruchus chinensis and Callosobruchus maculatus) in bulk stored chickpea and green gram. Bruchids hidden inside the grain kernels were detected acoustically through amplification and filtering of their mobility and feeding sounds. The multivariate technique of artificial neural network (ANN) was applied to assess and predict the bruchids’ density in bulk stored legumes. Five levels of bruchids density (0, 5, 10 15 and 20 bruchids per 500 g) were monitored under without insulation and with insulated condition on the basis of formant parameter obtained by analysis of the acoustic sensor data. The K fold validation method with back propagation multilayer perceptron methodology was used for the prediction of bruchids densities. The maximum and minimum values of accuracy (R2) of 0.99, 0.98 and 0.90, 0.89 could be achieved for both bruchids in stored green gram and chickpea under insulation and without insulation for the training and validation dataset, respectively. Least RMSE (0.82 and 0.89) was obtained for C. maculatus in sound insulated stored green gram for training and validation dataset, respectively. The accuracy of prediction and validation of experimental data with low RMSE and high R2 values for both the food legumes indicated that the ANN modeling performed well in predicting bruchids density. Hence it can be concluded that, best prediction was obtained for the C. maculatus for green gram under insulated condition. The results further corroborated that bioacoustic detection technique with ANN provided a reliable and accurate monitoring technique for bruchids. The developed technique can be adopted in large bulk storage grain systems for the selected legumes for predicting and assessing the growth of bruchids thereby leading to safer storage.

中文翻译：

---

通过生物声学检测和人工神经网络 (ANN) 评估散装鹰嘴豆和绿豆中的 bruchids 密度

摘要 为了评估散装鹰嘴豆和绿豆中食用豆科植物（Callosobruchus chinensis 和Callosobruchus maculatus）的生长和密度，利用昆虫取食和爬行声的声学检测，自动监测内部和外部谷物喂养的小黑麦。通过放大和过滤它们的移动和进食声音，从声学上检测到隐藏在谷粒内的 Bruchids。应用人工神经网络（ANN）的多元技术来评估和预测散装储存豆类中的毛刺密度。基于通过声学传感器数据分析获得的共振峰参数，在无绝缘和绝缘条件下监测五个级别的 bruchids 密度（0、5、10 15 和 20 个 bruchids/500 g）。带有反向传播多层感知器方法的 K 折验证方法用于预测 bruchids 密度。对于训练和验证数据集，存储的绿豆和鹰嘴豆在绝缘和没有绝缘的情况下，准确度 (R2) 的最大值和最小值分别为 0.99、0.98 和 0.90、0.89。分别在用于训练和验证数据集的隔音存储绿色革兰氏菌中获得了 C. maculatus 的最小 RMSE（0.82 和 0.89）。两种食物豆类的低 RMSE 和高 R2 值的实验数据的预测和验证的准确性表明，ANN 模型在预测 bruchids 密度方面表现良好。因此可以得出结论，在绝热条件下，对于绿革兰的C. maculatus 获得了最佳预测。结果进一步证实了带有 ANN 的生物声学检测技术为 bruchids 提供了可靠和准确的监测技术。所开发的技术可用于所选豆类的大型散装谷物储存系统，用于预测和评估 bruchid 的生长，从而实现更安全的储存。