(C) PLOS One [1]. This unaltered content originally appeared in journals.plosone.org.
Licensed under Creative Commons Attribution (CC BY) license.
url:https://journals.plos.org/plosone/s/licenses-and-copyright

------------



Individual dairy cow identification based on lightweight convolutional neural network

['Shijun Li', 'College Of Electronic', 'Information Engineering', 'Wuzhou University', 'Wuzhou', 'Lili Fu', 'College Of Information Technology', 'Jilin Agricultural University', 'Changchun', 'Yu Sun']

Date: 2021-12

In actual farms, individual livestock identification technology relies on large models with slow recognition speeds, which seriously restricts its practical application. In this study, we use deep learning to recognize the features of individual cows. Alexnet is used as a skeleton network for a lightweight convolutional neural network that can recognise individual cows in images with complex backgrounds. The model is improved for multiple multiscale convolutions of Alexnet using the short-circuit connected BasicBlock to fit the desired values and avoid gradient disappearance or explosion. An improved inception module and attention mechanism are added to extract features at multiple scales to enhance the detection of feature points. In experiments, side-view images of 13 cows were collected. The proposed method achieved 97.95% accuracy in cow identification with a single training time of only 6 s, which is one-sixth that of the original Alexnet. To verify the validity of the model, the dataset and experimental parameters were kept constant and compared with the results of Vgg16, Resnet50, Mobilnet V2 and GoogLenet. The proposed model ensured high accuracy while having the smallest parameter size of 6.51 MB, which is 1.3 times less than that of the Mobilnet V2 network, which is famous for its light weight. This method overcomes the defects of traditional methods, which require artificial extraction of features, are often not robust enough, have slow recognition speeds, and require large numbers of parameters in the recognition model. The proposed method works with images with complex backgrounds, making it suitable for actual farming environments. It also provides a reference for the identification of individual cows in images with complex backgrounds.

Introduction

Livestock farming is increasing in scale, informatization and refinement [1]. To achieve automated and informative daily management of large-scale cattle farms, individuals need to be able to be identified [2]. Traditionally, individual identification of cattle has been done by methods that cause permanent damage, such as engraved ear branding, ear tagging and radio frequency identification (RFID) tagging [3]. The ear carving method involves cutting openings in the ears, which is painful and time-consuming and can cause stress or even lead to violent death in severe cases [4]. Ear tags are often lost or damaged. Fosgate et al. [5] found that only 21% of range buffalo could be identified two years after ear tags were applied due to tag losses; therefore, tags are only a short-term solution. RFID technology uses radio waves for target identification and tracking and has become popular in dairy farms [6]. However, RFID systems have some security risks as they consist of tags, transponders and terminal servers and are prone to security problems such as tampering with tag contents, system crashes, and intrusion attacks on servers [7, 8]. At the same time, in the links that need real-time response, such as livestock production and disease prevention, it is difficult to apply this method to the actual cattle farm.

In recent years, there has been a trend towards using machine vision techniques to supervise the identification of cows, which has been led by rapid advances in deep learning [9]. Machine vision technology can achieve intelligent and accurate farming, and has the advantages of being low cost and reducing manual labour requirements. It is also non-contact and does not need to touch the animal for identification, so does not cause stress and can provide continuous long-term monitoring [10]. Many scholars have applied deep learning to cows [11, 12]. Zhao et al. [13] collected side-view videos of cows walking in a straight line to study and evaluate image processing techniques, including four feature extraction methods and two matching methods. The highest recognition accuracy of 96.72% was achieved when the FAST, SIFT and FLANN methods were used for feature extraction, description and matching, respectively. Zhang et al. [14] verified that their proposed deep convolutional network outperformed two traditional models, SIFT [15] and BOF [16], in recognizing individual cows in a farm environment. They conducted multiple comparative experiments with different network layers, convolutional kernel sizes and numbers of nodes in the fully connected layer. Shen et al. [17] used the YOLO model to detect cow targets in a series of side views of cows. They classified each cow by fine-tuning a convolutional neural network model and achieved 96.65% accuracy in individual cow identification. The studies above have used fixed side-view images of cows to train their network models; however, in real farms, cows are in constant motion and twisting of their bodies causes a certain degree of deformation of their side spots, which can influence identification and must be considered in practical applications.

Patrizia Tassinari et al. [18] used the yolov3 algorithm to identify cows moving in a cow pen. The cows’ rumps were used as the main detection area, and an average detection accuracy of 0.64–0.66 was obtained. Li et al. [19] proposed a convolutional neural network-based method for automated and accurate recognition of individual cows. It uses a residual learning inverse convolutional network to denoise cow images to obtain a training dataset. It improves on the InceptionV3 network to serve as a training master network and identifies individual cows from the patterns on their tails. Brahim Achour et al. [20] developed a non-invasive system based entirely on image analysis to identify individual cows and their behaviours based on the patterns on their heads. However, the small area of the head limits the available feature points, which has an impact on the final recognition results. Researchers such as Fumio Okura and Ran Bezen [21, 22] have used RGB-D camera 3D video analysis of cows for target detection and individual identification; however, the accuracy has much room for improvement. Although deep learning is widely used in the field of dairy cattle, the automated monitoring of individual cows is still in an early stage of research. Although some success has been achieved, many techniques cannot be applied to dairy farming in a generalized way [23].

In this study, we built a lightweight convolutional neural network model using the Alexnet model as a skeleton network. We trained the model with our own dataset obtained from a cattle farm, where cows were photographed under different lighting and pollution conditions against complex backgrounds, which are challenging to machine vision. In this paper, we demonstrate the effectiveness of our model on complex and variable datasets and provide a systematic analysis of its various modules.

The rest of the paper is organized as follows: Section 2 outlines the data acquisition and image sample expansion, and provides a background to the experimental method. Section 3 gives a detailed description of how the model was built and improved, then the experimental results are analysed in Section 4. Section 5 discusses the results and their limitations, and makes recommendations for future work.

[END]

[1] Url: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0260510

(C) Plos One. "Accelerating the publication of peer-reviewed science."
Licensed under Creative Commons Attribution (CC BY 4.0)
URL: https://creativecommons.org/licenses/by/4.0/


via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/