融合亮度自适应模块的端到端低光环境黑猪检测技术研究

doi:10.13304/j.nykjdb.2024.0493

摘要/Abstract

摘要：

为解决低光环境下黑猪图像质量差、识别与定位难度大，以及黑猪聚集时因遮挡和粘连导致的误检与漏检问题，提出LADnet（low-light animal detection network）检测模型。首先，采用亮度自适应模块（illumination-adaptive-transformer，IAT）及坐标注意力机制（coordinate attention，CA）对图像进行增亮与降噪处理；接着，设计了选择核卷积注意力（selective kernel convolutional attention，SKCA）模块，以提高模型对黑猪的感知能力；最后，利用ReLU激活函数解决了梯度消失和梯度爆炸等问题。结果表明，LADnet模型的精确率、召回率和平均准确率（mAP@0.5）分别达到97.32%、86.61%和92.73%，与基准模型相比，分别提升1.07、6.15和3.05个百分点；与单阶段目标检测模型SSD、YOLOv5相比，LADnet的平均准确率分别高出8.33、7.35个百分点；与两阶段目标检测模型Cascade R-CNN、Faster R-CNN、DAB_DETR相比，LADnet在检测精度提高的同时，具有更小的参数量和更快的检测速度，更符合实时检测要求。LADnet模型在低光黑猪检测任务中展现出卓越的检测性能和更高的鲁棒性，为低光环境下的黑猪精准识别提供了一种高效可靠的工具，对推动低光智慧养殖的发展具有重要意义。

关键词: 黑猪盘点, 目标检测, 注意力机制, 低光增强, 特征提取, YOLOv7, 智慧养殖

Abstract:

To address the issues of poor image quality， difficulty in recognition and localization， as well as false positives and false negatives caused by occlusion and adhesion in scenarios involving clustered black pigs under low-light conditions， a detection model named low-light animal detection network （LADnet） was proposed. Firstly， an illumination-adaptive transformer （IAT） and a coordinate attention （CA） mechanism were utilized to enhance the brightness and reduce noise in the images. Then， a selective kernel convolutional attention （SKCA） module was designed to improve the model’s ability to perceive black pigs. Finally， the ReLU activation function was employed to mitigate problems related to gradient vanishing and explosion. The results showed that the LADnet model achieved precision， recall and mean average precision （mAP@0.5） of 97.32%， 86.61% and 92.73%， respectively， representing improvements of 1.07， 6.15 and 3.05 percentage points compared to the baseline model. Compared to single-stage object detection models such as SSD and YOLOv5， LADnet achieved an average accuracy improvement of 8.33 and 7.35 percentage points， respectively. In comparison with two-stage models like Cascade R-CNN， Faster R-CNN and DAB_DETR， LADnet not only demonstrated higher detection accuracy but also achieved a smaller parameter size and faster detection speed， making it more suitable for the real-time detection requirements. The LADnet model demonstrated exceptional detection performance and enhanced robustness in low-light black pig detection tasks， providing an efficient and reliable tool for the accurate identification of black pigs in low-light environments， which holded significant importance for advancing the development of intelligent farming under low-light condition.

Key words: black pig inventory, object detection, attention mechanism, low-light enhancement, feature extraction, YOLOv7, intelligent breeding

中图分类号:

TP391.41

黄梦真, 李皞, 胡桓浚, 李梓芃, 盛钟尹, 刘义凡, 夏震言, 郑奥运. 融合亮度自适应模块的端到端低光环境黑猪检测技术研究[J]. 中国农业科技导报, 2025, 27(6): 113-125.

Mengzhen HUANG, Hao LI, Huanjun HU, Zipeng LI, Zhongyin SHENG, Yifan LIU, Zhenyan XIA, Aoyun ZHENG. Research on End-to-end Low-light Environment Black Pig Detection Technology Integrating IAT Module[J]. Journal of Agricultural Science and Technology, 2025, 27(6): 113-125.

图/表 13

图1 LADnet低光环境下黑猪检测网络结构

Fig. 1 LADnet black pig detection network structure in low light environment

图 2 IAT低光增强模块结构

Fig. 2 Structure of illumination-adaptive-transformer low light enhancement module

图 3 SKCA结构

Fig. 3 Structure of selective kernel convolutional attention

图 4 CA结构

Fig. 4 Structure of coordinate attention

图 5 R-SPPCSPC结构

Fig. 5 Structure of R-SPPCSPC

图6 增强算法效果图

Fig. 6 Enhanced algorithm effect diagram

表1 不同增强模块性能

Table 1 Performance of different enhancement modules

增强模块 Enhancement module	参数量 Params/K	平均精度（@0.5） mAP@0.5/%	平均精度（@0.50∶0.95） mAP@0.50∶0.95/%
Zero-DCE	6.5	89.32	51.45
Zero-DCE++	7.1	89.46	51.11
IAT	7.5	90.97	53.05
RUAS	56.8	89.61	52.62
URetinexNet	38.4	90.03	52.33

表2 各项改进策略的消融试验结果

Table 2 Results of ablation experiments with improved strategies

序号 No.	亮度自适应模块 IAT	激活函数 ReLU	坐标注意力机制 CA	选择核卷积注意力 SKCA	精确率 P/%	召回率 R/%	平均精度（@0.5） mAP@0.5/%	平均精度（@0.50∶0.95） mAP@0.50：0.95/%
1					96.25	80.46	89.68	52.78
2		√			95.79	80.98	90.60	52.83
3		√	√		95.98	83.03	90.85	53.08
4		√	√	√	95.25	84.68	91.37	53.88
5	√				95.80	83.64	90.97	53.05
6	√	√			96.61	82.70	91.50	53.40
7	√	√	√		96.80	83.80	91.80	54.50
8	√	√	√	√	97.32	86.61	92.73	54.84

图7 各项改进策略的热力图

Fig. 7 Heat map of each improved strategies

表3 不同目标检测算法的性能

Table 3 Performance of different object detection algorithms

模型 Model	平均精度（@0.5） mAP@0.5/%	平均精度（@0.50∶0.95） mAP@0.50∶0.95/%	每秒帧数 FPS	参数量 Params/M	计算量 FLOPs/G
SSD	84.40	41.10	35.3	23.75	30.43
Cascade R-CNN	86.00	47.40	34.2	69.15	205.24
Faster R-CNN	85.30	44.60	42.8	41.36	178.13
YOLOv5	85.38	43.16	175.4	21.24	49.02
YOLOv7	89.68	52.78	153.8	36.49	103.20
DAB_DETR	80.90	40.90	43.6	43.71	86.93
LADnet	92.73	54.84	129.8	40.58	116.90

图 8 模型改进前后检测效果注：方框所示为黑猪的分类标签；数字表示检测置信度，置信度越高，模型性能越好。

Fig. 8 Detection effects before and after model improvementNote：The box shows the classification label of the black pig；the number represents the detection confidence，and the higher the confidence， the better the model performance.

图9 模型改进前后训练可视化曲线

Fig. 9 Visual curves before and after model improvement

图 10 不同模型检测效果注：方框所示为黑猪的分类标签；数字表示检测置信度，置信度越高，模型性能越好。

Fig. 10 Detection effects of different modelsNote：The box shows the classification label of the black pig；the number represents the detection confidence，and the higher the confidence， the better the model performance.

参考文献 28

1	宋中山, 刘越, 郑禄,等.基于改进YOLOv3的自然环境下绿色柑橘的识别算法[J].中国农机化学报,2021,42(11):159-165.
	SONG Z S, LIU Y, ZHENG L, et al.. Identification of green citrus based on improved YOLOV3 in natural environment [J]. J. Chin. Agric. Mechan., 2021, 42(11):159-165.
2	LOEY M, MANOGARAN G, TAHA M H N,et al.. Fighting against COVID-19: a novel deep learning model based on YOLO-v2 with ResNet-50 for medical face mask detection [J/OL].Sustain. Cities Soc.,2021,65:102600 [2024-05-18]. .
3	王春梅,刘欢.YOLOv8-VSC:一种轻量级的带钢表面缺陷检测算法[J].计算机科学与探索,2024,18(1):151-160.
	WANG C M, LIU H.YOLOv8-VSC:lightweight algorithm for strip surface defect detection [J]. J. Front. Comput. Sci. Technol., 2024, 18(1):151-160.
4	伊建峰,黎思成,吕珊,等.基于频域数据增强及YOLOv7的动火作业检测模型[J].计算机应用,2023,43():285-290.
	YI J F, LI S C, LYU S, et al.. Hot work detection model based on frequency domain data enhancement and YOLOv7 [J]. J. Comp. Appl., 2023, 43(S2):285-290.
5	LIU W, ANGUELOV D, ERHAN D, et al.. Ssd: single shot multibox detector [C]// Computer Vision-ECCV 2016: 14th European Conference, Amsterdam, The Netherlands. Springer International Publishing, 2016: 21-37.
6	GIRSHICK R. Fast R-CNN [C]// Proceedings of the IEEE International Conference on Computer Vision. IEEE, 2015: 1440-1448.
7	REN S, HE K, GIRSHICK R, et al.. Faster R-CNN: towards real-time object detection with region proposal networks [J].IEEE Trans. Pattern Anal. Mach. Intell., 2017, 39(6): 1137-1149.
8	ZHENG C, ZHU X M, YANG X F, et al.. Automatic recognition of lactating sow postures from depth images by deep learning detector [J]. Comput. Electron. Agric., 2018, 147:51-63.
9	AHN H, SON S, KIM H, et al.. EnsemblePigDet: ensemble deep learning for accurate pig detection [J/OL]. Appl. Sci., 2021, 11(12): 5577 [2024-05-18]. .
10	HUANG L, XU L, WANG Y, et al.. Efficient detection method of pig-posture behavior based on multiple attention mechanism [J/OL].Comput. Intell. Neurosci., 2022, 2022: 1759542 [2024-05-18]..
11	LEINGARTNER M, MAURER J, FERREIN A, et al.. Evaluation of sensors and mapping approaches for disasters in tunnels [J]. J. Field. Robot., 2016, 33(8): 1037-1057.
12	XIAO Y, JIANG A, YE J, et al.. Making of night vision: object detection under low-illumination [J]. IEEE Access, 2020, 8: 123075-123086.
13	XU X, WANG S, WANG Z, et al.. Exploring image enhancement for salient object detection in low light images [J]. ACM Trans. Multimedia Comput. Commun. Appl., 2021, 17(1s): 1-19.
14	ZHANG S, TUO H, HU J, et al.. Domain adaptive yolo for one-stage cross-domain detection [C]// Proceedings of Asian Conference on Machine Learning. PMLR, 2021: 785-797.
15	WANG J, YANG P, LIU Y, et al.. Research on improved yolov5 for low-light environment object detection [J/OL]. Electronics, 2023, 12(14): 3089 [2024-05-18]. .
16	QIU Y, LU Y, WANG Y, et al.. IDOD-YOLOV7:image-dehazing YOLOV7 for object detection in low-light foggy traffic environments [J/OL].Sensors (Basel Switz.),2023,23(3):1347[2024-05-18]. .
17	WANG C Y, LIAO H Y M, WU Y H, et al.. CSPNet: a new backbone that can enhance learning capability of CNN [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. IEEE, 2020: 390-391.
18	HOU Q, ZHOU D, FENG J. Coordinate attention for efficient mobile network design [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2021: 13713-13722.
19	GUO C, LI C, GUO J, et al.. Zero-reference deep curve estimation for low-light image enhancement [C]// Proceedings of the IEEE/CVF Conference on Computer Vvision and Pattern Recognition. IEEE, 2020: 1780-1789.
20	LI C, GUO C, LOY C C.Learning to enhance low-light image via zero-reference deep curve estimation [J]. IEEE Trans. Pattern Anal. Mach. Intell., 2022, 44(8):4225-4238.
21	CUI Z, LI K, GU L, et al.. You only need 90K parameters to adapt light: a light weight transformer for image enhancement and exposure correction [J/OL]. 2022, 5: 14871 [2024-05-18]. .
22	LIU R S, MA L, ZHANG J A,et al..Retinex-inspired unrolling with cooperative prior architecture search for low-light image enhancement [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2021: 10561-10570.
23	WU W, WENG J, ZHANG P, et al.. Uretinex-net: Retinex-based deep unfolding network for low-light image enhancement [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2022: 5901-5910.
24	WOO S, PARK J, LEE J Y,et al.. CBAM:convolutional block attention module [C]// Proceedings of the European Conference on Computer Vision (ECCV). 2018: 3-19.
25	LI X, WANG W, HU X, et al.. Selective kernel networks [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2019: 510-519.
26	HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2018: 7132-7141.
27	HAYOU S, DOUCETt A, ROUSSEAU J. On the impact of the activation function on deep neural networks training [C]// International Conference on Machine Learning. PMLR, 2019: 2672-2680.
28	JIANG P, ERGU D, LIU F, et al.. A review of Yolo algorithm developments. [J]. Proc. Comput. Sci., 2022, 199: 1066-1073.

[1]	袁嘉良, 连润楠, 张吴平. 黄花菜花蕾的精准识别与分级方法[J]. 中国农业科技导报, 2025, 27(5): 103-112.
[2]	熊晓菲, 王秀琴, 庄翠珍, 郭家贤, 谢新锐, 吴建伟, 李奇峰. 基于ROI融合特征的柑橘炭疽病诊断方法[J]. 中国农业科技导报, 2024, 26(9): 83-92.
[3]	帖军, 赵捷, 郑禄, 吴立锋, 洪博文. 改进YOLOv5 模型在自然环境下柑橘识别的应用[J]. 中国农业科技导报, 2024, 26(7): 111-120.
[4]	李锐风, 杨云福, 杨永发, 于永顺. 基于机器视觉的玫瑰花检测与特征提取[J]. 中国农业科技导报, 2024, 26(4): 106-113.
[5]	李名博, 刘玉乐, 穆志民, 郭俊旺, 卫勇, 任东悦, 贾济深, 卫泽中, 栗宇红. 基于YOLOX-L-TN模型的番茄果实识别[J]. 中国农业科技导报, 2024, 26(4): 97-105.
[6]	张宏, 李卫国, 张晓东, 卢必慧, 张琤琤, 李伟, 马廷淮. 基于HJ-1星和GF-1号影像融合特征提取冬小麦种植面积[J]. 中国农业科技导报, 2024, 26(2): 109-119.
[7]	黄诗锐, 王天一, 文韬, 周江龙. 基于改进YOLOv7的农作物虫害识别[J]. 中国农业科技导报, 2024, 26(11): 107-116.
[8]	彭明康, 崔钰, 薛淇元, 殷允振, 尹哲, 张吴平, 李富忠. 基于天气数据增强和微调的苗期作物杂草识别定位模型[J]. 中国农业科技导报, 2024, 26(10): 125-134.
[9]	赵越, 卫勇, 单慧勇, 穆志民, 张健欣, 吴海云, 赵辉, 胡建龙. 基于深度学习的高分辨率麦穗图像检测方法[J]. 中国农业科技导报, 2022, 24(9): 96-105.
[10]	石瑞瑶,田有文*,赖兴涛,古文君. 基于机器视觉的苹果品质在线分级检测[J]. 中国农业科技导报, 2018, 20(3): 80-86.
[11]	蒋艳娜1,肖伯祥2,郭新宇1,2,杨宝祝1,2*. 基于视频的植物动画合成方法[J]. , 2015, 17(1): 115-121.