Industry 4.0 is currently considered the structural implementation of networked and cooperative digitalisation and the next step in technological and social development. The aim of this paper is to examine how these structures are also suitable for agriculture and whether there are already approaches to this. Therefore, the main aspects of Industry 4.0 will be analysed and compared with agricultural examples from arable farming and livestock farming. The study shows that the approaches of Industry 4.0 are also useful for agriculture. However, they must be adapted to agriculture, as it has a different basic structure. As in industry, it is also evident in agriculture that there is still a need for action in the organisational and technical networking of systems.
The term ”Agriculture 4.0” emerged from the term “Industry 4.0” like amany other “4.0” terms. However, are Industry 4.0 technologies and concepts really applicable to agriculture? Are the benefits that Industry 4.0 brings to industrial use cases transferable to livestock farming? This paper tries to answer this question for the three dominant sectors of livestock farming in Central Europe and Germany: Poultry, pig fattening, and dairy farming. These sectors are analyzed along with the eight most relevant Industry 4.0 benefits. The results show that only part of the Industry 4.0 benefits are relevant for livestock farming in a similar manner as in industrial production. Due to basic differences between industrial and livestock farming use cases, some of the benefits must be adapted. The presence of individual living animals and the strong environmental impact of livestock farming affect the role of digital individualization and demand orientation. The position of livestock farming within the value chain minimizes the need for flexibilization. The introduction and adoption of Industry 4.0 concepts and technologies may contribute significantly to transforming agriculture into something that may be called Agriculture 4.0. Technologies are indispensable for this development step, but vocational education and open-mindedness of farmers towards Industry 4.0 is essential as well.
The particularities of the agricultural ecosystem of cooperating partners suggest a highly distributed, open data architecture to support the exchange and re-use of data and data-driven applications. We describe the data strategy and the federated basic architecture of the Agri-Gaia ecosystem. We present an ontology-based approach to metadata management which extends a bucket store for arbitrary data storage by an RDF-based metadata graph store and employs widely-used domain ontologies as a conceptual basis. Data and service providers are free to extend the describing metadata at any time, according to their needs. The resulting set of interconnected platforms supports the publication, retrieval and secure sharing of exposable data under full control of their owners.
The task of re-identifying (ReID) a previously seen object from a different camera angle is a rising research topic in the computer vision community. Most work is done in the domain of person re-identification, though. In this paper we present the applicability of two novel datasets for animal ReID on algorithms for person ReID, which is an unexamined topic today. The first dataset consists of images of 376 different wild common toads in many different poses and camera orientations. The second dataset contains smaller amounts of images showing different species of zoo animals. We evaluated a variety of different ReID algorithms on both datasets. The results show that in our setting even previously unseen toads can be re-identified with a Rank-1 score greater than 93% using the already established algorithms of person ReID. Interestingly, we found that a mixed dataset containing different animal species increases the identification performance for the single species task in comparison to a per-species training.
This empirical study aims at improving the effectiveness of the standard 2-layer MLP projection head $g(\cdot)$ featured in the SimCLR framework through the use of pretrained autoencoder embeddings. Given a contrastive learning task with a largely unlabeled image classification dataset, we first train a shallow autoencoder architecture and extract its compressed representations contained in the encoder's embedding layer. After freezing the weights within this pretrained layer, we use it as a drop-in replacement for the input layer of SimCLR's default projector. Additionally, we also apply further architectural changes to the projector by decreasing its width and changing its activation function. The different projection heads are then used to contrastively train and evaluate a feature extractor $f(\cdot)$ following the SimCLR protocol, while also examining the performance impact of Z-score normalized datasets. Our experiments indicate that using a pretrained autoencoder embedding in the projector can not only increase classification accuracy by up to 2.9% or 1.7% on average but can also significantly decrease the dimensionality of the projection space. Our results also suggest, that using the sigmoid and tanh activation functions within the projector can outperform ReLU in terms of peak and average classification accuracy. When applying our presented projectors, then not applying Z-score normalization to datasets often increases peak performance. In contrast, the default projection head can benefit more from normalization. All experiments involving our pretrained projectors are conducted with frozen embeddings, since our test results indicate an advantage compared to using their non-frozen counterparts.