太阳能杀虫灯作为一款绿色无污染的植保器械，被喻为是农田“守卫”，通过金属网释放的高压脉冲电流能够有效的消灭趋光性害虫，给农业生产提供保障。随着现代农业的发展，太阳能杀虫灯单一的被应用于消灭害虫已经不能满足农业生产需求，更加重要的是通过计算杀死害虫的数量进一步掌握杀虫灯覆盖农田区域的害虫密度。一方面对害虫密度的统计可以起到虫害监测的作用，当害虫密度过高时需要采取对应的措施，降低损失。另外一方面，太阳能杀虫灯部署在不同的农田区域，可以结合杀虫数量掌握不同农田区域的害虫密度，针对不同农田区域的害虫密度实施不同剂量的农药，以减少农药滥用。基于上述分析，掌握农田区域的害虫密度至关重要，而统计害虫密度的关键在于如何精确的计算出杀虫灯杀死害虫的数量。现有的太阳能杀虫灯杀虫计数方法为电压计数，该方法是将高压网放电电击害虫时产生的电压脉冲数量认作为杀虫数量，当体积较大害虫黏住在高压网时很容易造成重复计数。本文将从传统的电压计数方法缺点分析，综合对比杀虫放电时的电压和声音特征，研究出一种能够实现精确统计杀虫灯杀虫害虫数量的方法。本文的研究内容主要包括以下几个方面：

（1）构建太阳能杀虫灯杀虫计数数据集。本文分别在校园的实验楼阳台和校园室外搭建杀虫灯硬件平台，在杀虫灯上安装具备视频录制及储存功能的摄像头录制杀虫视频，通过电压比较器采集杀虫放电时产生的电压脉冲数据，制作成形式为电压、声音和杀虫数量一一对应的杀虫计数数据集。

（2）综合分析杀虫数据。杀虫数据包括杀虫放电电压数据、杀虫放电声音数据和杀虫数量。对电压数据分析后，实验发现电压数据容易重复计数的原因是一些体积较大的害虫难以被一次电击致死，不断产生电压数据导致的。提取杀虫声音的时域和频域特征，使用互相关分析验证了不同类型杀虫声音之间是具有高度相似性的；基于此，初步设计了利用声音计数的方法，采用双门限端点检测的方法，检测了音频数据杀虫声音片段的起止时间，分析了相邻杀虫脉冲时间之间的差值分布情况，为后续杀虫计数提供依据。最后使用变分模态分解对声音进行分解，去除第一级和第二级本征模态后重构杀虫声音，实现了对声音质量的增强，增强之后的信噪比为5.46dB。

（3）对比不同杀虫计数方法的实验结果。实验按照电压脉冲数值和杀虫数量是一一对应的关系，即杀死一只害虫产生一个电压脉冲，在实验的数据集上验证了电压计数方法容易重复计数。根据杀虫声音分析的结果，本文设计了一种利用相邻声音脉冲时间差值间隔计数的方法，实验结果表明当时间差值设置为0.22s时，可以在符合实际情况的同时实现精确计数，声音计数方法可以减少由电压数据容易带来重复计数问题。最后结合电压和声音计数各自的优点，使用机器学习的方法结合两者结果训练学习得到最终杀虫计数结果，在校园数据验证集得到了90.02%准确率，比单独使用电压数据计数（78.78%）提高了11.24%，该结果表明利用电压和声音联合计数的方法能够实现精确计数。

（4）方法应用的泛化性验证。为了验证本文方法在害虫数量多的农田环境依然可以适用，实验在八卦洲农业实验基地部署杀虫灯并采集杀虫计数数据，并用在校园实验得到的方法进行计数验证，在八卦洲验证数据上得到了93.28%的准确率，说明了本文方法在农田环境依旧适用，且在声音计数设置的0.22s阈值时间也适用于农田环境。最后本文用电脑作为声音采集平台和方法验证平台，通过人工模拟害虫放电的过程，对放电的声音录取、保存、降噪、端点检测和加载电压数据机器学习计数操作，验证了方法在硬件平台移植的可行性。

Solar insecticidal lamp (SIL), as a green and pollution-free plant protection equipment, is regarded as the "guardian" of farmland. The high-voltage pulse current released through metal mesh can effectively eliminate phototactic pests and provide a guarantee for agricultural production. With the development of modern agriculture, the single application of SIL to eliminate pests can no longer meet the needs of agricultural production. It is more important to further grasp the pest density in the farmland covered by SIL by calculating the number of pests killed. On the one hand, the statistics of pest density can play the role of pest monitoring. When the pest density is too high, corresponding measures should be taken to reduce the loss. On the other hand, SILs are deployed in different farmland areas, which can grasp the pest density in different farmland areas according to the insecticidal quantity, and implement different doses of pesticides according to the pest density in different farmland areas, which can reduce pesticide abuse. Based on the above analysis, it is very important to master the density of pests in farmland areas, and the key to statistics of pest density is how to accurately calculate the number of pests killed by SIL. The existing counting method of SIL is voltage counting, which regards the number of voltage pulses generated when the high-voltage network discharges electric shocks to pests as the number of pests, and it is easy to cause repeated counting when large pests stick to the high-voltage network. In this paper, the shortcomings of the traditional voltage counting method are analyzed, and the voltage and sound characteristics of insect discharge are comprehensively compared, and a method that can accurately count the number of insect pests by insect lamp is developed. The research contents of this paper mainly include the following aspects:

(1) Constructing the dataset of SIL insecticidal counting. In this paper, the hardware platform of insecticidal lamp was built on the balcony of the campus experimental building and outside the campus, and a camera with video recording and storage function was installed on the insecticidal lamp to record the insecticidal video. The voltage pulse data generated during insecticidal discharge was collected by a voltage comparator, and the insecticidal counting dataset was made in the form of one-to-one correspondence between voltage, sound and insecticidal quantity.

(2) Comprehensive analysis of insecticidal data. Insecticidal data include insecticidal discharge voltage data, insecticidal discharge sound data and insecticidal quantity. After analyzing the voltage data, it is found that the reason why the voltage data are easy to be counted repeatedly is that some large pests are difficult to be killed by one electric shock and constantly produce voltage data. The time domain and frequency domain characteristics of insecticidal sounds are extracted, and the high similarity between different types of insecticidal sounds is verified by cross-correlation analysis. Based on this, the method of sound counting and double-threshold endpoint detection are preliminarily designed, and the start and end times of audio data insecticidal sound segments are detected, and the difference distribution between adjacent insecticidal pulse times is analyzed, which provides a basis for subsequent insecticidal counting. Finally, the sound is decomposed by variational modal decomposition, and the insecticidal sound is reconstructed after removing the first and second eigenmodes, which realizes the enhancement of sound quality, and the signal-to-noise ratio after enhancement is 5.46dB.

(3) The experimental results of different insecticidal counting methods were compared: According to the one-to-one relationship between the voltage pulse value and the number of pesticides, that is, killing a pest produces a voltage pulse, which proves that the voltage counting method is easy to count repeatedly on the experimental dataset. According to the results of insecticidal sound analysis, this paper designs a method of interval counting by using the time difference between adjacent sound pulses. The experimental results show that when the time difference is set to 0.22s, accurate counting can be realized while conforming to the actual situation, and the sound counting method can reduce the problem of repeated counting easily caused by voltage data. Finally, combining the respective advantages of voltage and sound counting, the machine learning method is used to train and learn the results of both, and the final insecticidal counting result is obtained. In the campus data verification set, the accuracy rate is 90.02%, which is 11.24% higher than that of voltage data counting alone (78.78%). The result shows that the method of voltage and sound combined counting can realize accurate counting.

(4) Generalization verification of method application: In order to verify that the method in this paper can still be applied in the farmland environment with a large number of pests, the experiment deployed SIL to collect insecticidal counting data in Baguazhou agricultural experimental base, and the method obtained in the campus experiment was used for counting verification, and the accuracy rate of the verification data in Baguazhou was 93.78%, which showed that the method in this paper was still applicable in the farmland environment, and the threshold time of 0.22s set by sound counting was also applicable to the farmland environment. Finally, this paper uses the computer as the sound collection platform and the method verification platform, and verifies the feasibility of transplanting the method to the hardware platform by artificially simulating the process of insect discharge, and performing machine learning counting operations on the sound recording, storage, noise reduction, endpoint detection and voltage data loading.

[1] 安宁生物科技有限公司. 太阳能杀虫灯: 型号AN-D-1006[EB/OL].

http://ajansw.com/detail-201618-81106.html.

[2] 蔡展标, 姜晟, 黄杰贤. 基于4G网络的电击式杀虫测报灯[J]. 嘉应学院学报, 2021, 39(06): 10-18.

[3] 陈彬, 刘燕, 谭本垠, 等. 一种太阳能杀虫灯光控雨控时控功能检测系统及检测方法: 中国, 202211553960.1[P]. 2023-07-07.

[4] 陈世雄. 基于单片机的杀虫灯智能控制器设计[J]. 江苏农业科学, 2021, 49(13): 186-191.

[5] 成都比昂科技有限公司. 太阳能杀虫灯: 型号BA-T/RL[EB/OL].

https://www.cdbeyond.com/a/104.html.

[6] 成都瑞昌仪器制造有限公司. 太阳能杀虫灯: 型号RC-NX-LP312S[EB/OL].

https://www.cdbeyond.com/a/104.html.

[7] 葛轶洲, 许翔, 杨锁荣, 等.序列数据的数据增强方法综述[J]. 计算机科学与探索,2021,15(07):1207-1219.

[8] 韩云霄, 邵清, 符玉襄, 等. 复杂噪声中基于MFCC距离的语音端点检测算法[J]. 计算机工程, 2020, 46(3): 309-314.

[9] 郝雪丽, 李会宾, 段玉林, 等. 基于SAM的田块提取方法在田块平整成效评价中的应用研究[J]. 中国农业信息, 2023, 35(05): 1-10.

[10] 何海洋. 一种杀虫灯电网触电害虫计数系统和统计方法: 中国, 201711297675.7[P]. 2018-06-08.

[11] 河北一航新能源科技有限公司. 太阳能杀虫灯: 型号HSC-01 [EB/OL].

http://www.insecticidallamp.cn/product/744.html/

[12] 河南格润利格农林设备有限公司. 太阳能杀虫灯: 型号RY-15II-1[EB/OL].

http://www.hngrlg.com/product/shachongdeng/zldscd/209.html.

[13] 胡古月, 汪小旵, 鲁伟, 等. 基于Hilbert变换的害虫自动计数和发布系统的设计[J]. 湖南农业大学学报(自然科学版), 2017, 43(01): 103-107.

[14] 纪振发, 杨晖, 李然, 等. 基于短时自相关及过零率的语音端点检测算法[J]. 电子科技, 2016 ,29(09): 52-55.

[15] 李凯亮, 舒磊, 黄凯, 等. 太阳能杀虫灯物联网研究现状与展望[J]. 智慧农业, 2019, 1(03): 13-28.

[16] 李耀明，梁振伟，徐立章. 一种基于碰撞声音识别的籽粒计数传感器: 中国, 201410293237.3[P]. 2017-04-12.

[17] 林叶蓁, 林峰. 一种基于声音信息的跳绳计数方法: 中国, 201910233228.8[P]. 2020-12-01.

[18] 刘慧, 杨玉清, 鲁梦瑶, 等. 农产品市场环境下语音端点检测的研究[J]. 自动化技术与应用, 2019, 38(05): 129-134.

[19] 刘万才, 刘振东, 黄冲, 等. 近10年农作物主要病虫害发生危害情况的统计和分析[J]. 植物保护, 2016, 42(05): 1-9+46.

[20] 路敬祎, 马雯萍, 叶东, 等. 基于VMD的声音信号增强算法研究[J]. 机械工程学报, 2018, 54(10): 10-15.

[21] 马庆禄, 邹政, 刘丰杰. 基于行车声音端点检测的交通量统计[J]. 科学技术与工程, 2020, 20(04): 1676-1683.

[22] 明莉, 胡伟. 电子灯光诱杀虫技术发展[J]. 农业工程, 2012, 2(S1): 11-15.

[23] 欧莉莎. 土壤有机磷农药污染及其修复方法研究进展[J]. 乡村科技, 2022, 13(23): 156-158.

[24] 邱荣洲, 赵健, 池美香, 等. 基于物联网的害虫智能监测系统设计与实现[J]. 福建农业学报, 2020, 35(02): 235-242.

[25] 汪思梦, 田中旭, 鲁洪江. 改进小波包降噪与RSSD的滑动轴承声音诊断方法[J]. 机械设计与研究, 2023, 39(03): 121-125.

[26] 王彦祖, 李白. 农药污染对我国生态环境的影响及对策分析[J]. 皮革制作与环保科技, 2022, 3(15): 77-79.

[27] 王永华, 刘佳, 董志平, 等. 一种多功能杀虫灯系统及其杀虫方法: 中国, 202310495998.6[P]. 2023-07-04

[28] 吴新忠, 夏令祥, 张旭, 等. 基于谱熵梅尔积的语音端点检测方法[J]. 北京邮电大学学报, 2019, 42(02): 83-89.

[29] 肖德琴, 招胜秋, 刘又夫, 等. 一种基于轻量级神经网络的计数杀虫灯及计数方法: 中国, 202210843230.9 [P]. 2023-09-26.

[30] 新乡市天意新能源科技开发有限公司. 太阳能杀虫灯: 太阳能电击式杀虫灯[EB/OL].

http://xxtyxny.com/proinfo/21.html.

[31] 徐练, 文礼章. 影响杀虫灯诱虫效果的因素及其发展方向[J]. 中国植保导刊, 2015, 35(05): 19-22.

[32] 徐瑞清, 杨文丽, 胡桂兰，等. 农用杀虫灯在我国的研发及应用分析[J]. 中国农业文摘-农业工程, 2022, 34(01): 68-70.

[33] 扬州市宝迪照明科技有限公司. 太阳能杀虫灯: 太阳能频振式杀虫灯[EB/OL].

https://www.yzbaodi.com/newview3834.html.

[34] 张凤荣. 分区分类分时序建设高标准农田[J]. 中国农业综合开发, 2023(04): 21-24.

[35] 张锦芳, 张阳, 徐文平, 等. 昆虫的趋光性及其应用于害虫治理的研究进展[J].世界农药, 2020, 42(11): 26-35.

[36] 赵秀花. 蔬菜害虫抗药性及防治对策[J]. 新农业, 2022(17): 27-28.

[37] 浙江隆皓农林科技有限公司. 太阳能杀虫灯: 型号TDB-2013[EB/OL].

http://www.dwdds.com/products_detail/c-_detailId%3D1171498953059016704.html.

[38] 浙江托普云农科技股份有限公司. 太阳能杀虫灯: 型号TPSC3-2[EB/OL].

https://www.tpyn.net/productshow68.html.

[39] 中华人民共和国农业农村部. 农业农村部办公厅关于印发《2024年“虫口夺粮”保丰收行动方案》的通知 [EB/OL]. 2024-03-07.

http://www.moa.gov.cn/govpublic/ZZYGLS/202403/t20240314_6451424.htm.

[40] 朱克凡, 王杰贵, 刘有军. 小样本条件下基于数据增强和WACGAN的雷达目标识别算法[J]. 电子学报, 2020, 48(06): 1124-1131.

[41] 卓嘎, 边巴旺堆, 姜军. 基于短时平均能量和短时过零率的藏语语音端点检测研究[J]. 电脑知识与技术, 2014, 10(31) :7466-7469.

[42] A Valipour, R Abbasi-Kesbi. A heartbeat and respiration rate sensor based on phonocardiogram for healthcare applications[C]. 2017 Iranian Conference on Electrical Engineering (ICEE), Tehran, Iran, 2017, pp. 45-48.

[43] Ahmed G, Lawaye A A. CNN-based speech segments endpoints detection framework using short-time signal energy features[J]. International Journal of Information Technology, 2023, 15(8): 4179-4191.

[44] Ahsan Adeel, Mandar Gogate, Amir Hussain. Contextual deep learning-based audio-visual switching for speech enhancement in real-world environments[J]. Information Fusion, volume 59, 2020, pages 163-170, ISSN 1566-2535

[45] C Donahue, B Li, R Prabhavalkar. Exploring speech enhancement with generative adversarial networks for robust speech recognition[C]. 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Calgary, AB, Canada, 2018, pp. 5024-5028.

[46] Cao D, Gao X, Gao L. An improved endpoint detection algorithm based on MFCC Cosine Value[J]. Wireless personal communications, 2017, 95: 2073-2090.

[47] Cubuk E D, Zoph B, Mane D, et al. Autoaugment: Learning augmentation policies from data[J]. arxiv preprint arxiv: 1805. 09501. 2018.

[48] D Emmanouilidou, E D McCollum, D E Park, et al. Computerized lung sound screening for pediatric auscultation in noisy field environments[J]. IEEE Transactions on Biomedical Engineering, vol. 65, no. 7, pp. 1564-1574, July 2018.

[49] David C Kazyak, Amy M Flowers, Nathan J Hostetter. Integrating side-scan sonar and acoustic telemetry to estimate the annual spawning run size of Atlantic sturgeon in the Hudson River[J]. Canadian Journal of Fisheries and Aquatic Sciences. 2020, 77(6): 1038-1048.

[50] Dragomiretskiy K, Zosso D. Variational mode decomposition[J]. IEEE transactions on signal processing, 2013, 62(3): 531-544.

[51] Eliopoulos P A, Potamitis I, Kontodimas D C, et al. Detection of adult beetles inside the stored wheat mass based on their acoustic emissions[J]. Journal of economic entomology, 2015, 108(6): 2808-2814.

[52] Hao X, Shan C, Xu Y, et al. An attention-based neural network approach for single channel speech enhancement[C]. ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, UK, 2019, pp. 6895-6899.

[53] Hou J, Wang S, Lai Y, et al. Audio-Visual speech enhancement using multimodal deep convolutional neural networks[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 2, no. 2, pp. 117-128, April 2018.

[54] Huang K, Shu L, Li K, et al. A Novel Accurate Insecticidal Counting Method Based on Solar Insecticidal Lamp using Machine Learning[J]. IEEE Transactions on Consumer Electronics (2023).

[55] Huang K, Shu L, Li K, et al. Insecticidal counting dataset based on one solar insecticidal lamp and two cameras. Frontiers in Plant Science, 2022, 13: 995118.

[56] I Hwang, J H Chang. End-to-End speech endpoint detection utilizing acoustic and language modeling knowledge for online low-latency speech recognition[J]. IEEE Access, vol. 8, pp. 161109-161123, 2020.

[57] IWANA B K, Uchida S. An empirical survey of data augmentation for time series classification with neural networks[J]. Plos one, 2021, 16(7): e0254841.

[58] Kim K N, Huang Q Y, Lei C L. Advances in insect phototaxis and application to pest management: A review[J]. Pest management science, 2019, 75(12): 3135-3143.

[59] Li Y, Quan H, Xiao C. Speech enhancement using ICA with EMD[J]. Journal of Central China Normal University, 2015, 49(1).

[60] N Yella, B Rajan. Data augmentation using GAN for sound based COVID 19 diagnosis[C]. 2021 11th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS), Cracow, Poland, 2021, pp. 606-609.

[61] Nazari M, Sakhaei S M. Successive variational mode decomposition[J]. Signal Processing, 2020, 174: 107610.

[62] Oh Cheolhwan, Seungmin Han, Jongpil Jeong. Time-series data augmentation based on interpolation[J]. Procedia Computer Science 175 (2020): 64-71.

[63] P Mowlaee, C Nachbar. Speaker dependent speech enhancement using sinusoidal model[C]. 2014 14th International Workshop on Acoustic Signal Enhancement (IWAENC), Juan-les-Pins, France, 2014, pp. 80-84.

[64] Q Wu, Y Liu. A speech endpoint detection method based on cascaded speech enhancement[C]. 2021 International Conference on Electronic Information Technology and Smart Agriculture (ICEITSA), Huaihua, China, 2021, pp. 1-6.

[65] Ray P P. Internet of things for smart agriculture: Technologies, practices and future direction[J]. Journal of Ambient Intelligence and Smart Environments, 2017, 9(4): 395-420.

[66] Sebastián-González E, Camp R J, Tanimoto A M, et al. Density estimation of sound-producing terrestrial animals using single automatic acoustic recorders and distance sampling[J]. Avian Conservation and Ecology, 2018.

[67] Smith J A, Holt K, Dockry R, et al. Performance of a digital signal processing algorithm for the accurate quantification of cough frequency[J]. European Respiratory Journal, 2021, 58(2).

[68] Tao T, Zheng H, Yang J, et al. Sound localization and speech enhancement algorithm based on dual-microphone[J]. Sensors 2022, 22, 715.

[69] Telaumbanua M, Haryanto A, Wisnu F K, et al. Design of insect trap automatic control system for cacao plants[J]. Procedia Environmental Science, Engineering and Management, 2020, 8(1): 167-175.

[70] Tu H, Tang N, Hu X, et al. LED multispectral circulation solar insecticidal lamp application in rice field[J]. Transactions of the Chinese Society of Agricultural Engineering, 2016, 32(16): 193-197.

[71] Wang L, Han M, Li X, et al. Review of classification methods on unbalanced dataset. IEEE Access 9 (2021): 64606-64628.

[72] Yao H, Shu L, Yang F, et al. The phototactic rhythm of pests for the solar insecticidal lamp: A review[J]. Frontiers in Plant Science, 2023, 13: 1018711.

[73] Zeng Q, Liu J, Yang D, et al. Machine Learning Based Automatic Sport Event Detection and Counting[C]. 2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC), Beijing, China, 2021, pp. 16-20.

[74] Zhao C. State-of-the-art and recommended developmental strategic objectivs of smart agriculture[J]. Smart Agriculture, 2019, 1(1): 1-7.