Food security is a major problem worldwide. Ensuring that the crops produced are both safe and wholesome is crucial not only for people as the ultimate consumers of the crops, but also for farmers. Plant diseases are responsible for a significant percentage of crop losses. This alleviates the need for a fast and accurate model to discriminate and identify plants with diseases. The chapter aims to achieve the same through deep learning. The data set used in the work was obtained from Plant Village Dataset. The work customs deuce pre-trained models, EfficientNetB0 and DenseNet121, to citation the traits of the plants. The extracted traits are then fused together through concatenation to allow the model to read the more meaningful crop trait data. This also ensures that the different sets of feature data read by the two models compensate for any feature loss during extraction. It turns out that the above method gives better results associated to other models.
Top1. Introduction
Food Security indicates the availability of food at all times and the accessibility of individuals to it. According to an article published in ‘Agronomy’- factors that influence crop yield and hence food security can be grouped into three categories namely technological, biological and environmental. The amount of influence of each of these factors on crop production again varies based on region to region and crop to crop. In these factors plant diseases make up a major part of the biological group with pests, weeds and insects. It is estimated by FAO that annually the loss due to pests is up to 40% of global crop production. Every year, the global economy is affected by over $220 billion due to plant diseases and a minimum of $70 billion due to invasive insects. Especially in a highly populated country like India, which is one of the largest producers of most food crops around the world, the threat to food security is even more imminent. Hence it is imperative that new and better methods to identify plant diseases are explored.
Deep Learning, a subset of machine learning, is in use across the various industries ranging from driving to medical devices. Deep learning at it’s root is a neural network with layers. These layers attempt to replicate the neural functioning of the human brain to enable it to process and learn from large amounts of data. One of the key recompences of using deep learning algorithms apart from the large amounts of data it can work with is the kind of data it can work with. Deep Learning algorithms cannister progress unstructured data, for example text and images. It mechanizes feature extraction reducing the involvement of humans and also performs pattern analysis and data classification.
The traditional methods of detecting plant diseases involve visual inspection and later detailed detection in the laboratory. This process is both time-consuming and accessible to farmers from different economic backgrounds, such as B. small farmers, not accessible. This involves the development of automated and intelligent illness detection systems that deliver faster findings and are more accessible through the use of diverse technologies such as artificial intelligence, machine learning, and deep learning. Deep learning techniques gather features from pictures and then utilise these functions to conduct classification or regression as needed.
The proposed work is a unique classification model to correctly categorize corn leaves from digital images of gray spot, leaf rust, northern leaf blight and healthy leaves. This study uses two pre-trained deep convolutional neural networks (DCNNs), EfficientNetB0 and DenseNet121. Appropriate parameter selections and feature fusion techniques are used to combine the predictive capabilities of the two models and create an end-to-end classification modelOur primary contributions are as follows:
- •
Characteristics are retrieved from two separate models and merged.
- •
A appropriate amount of parameters increases classification accuracy.
- •
A graphical user interface (GUI) for a deep learning model for recognising and categorising maize leaf plant diseases is suggested.
Top2. Literature Survey
Deep learning algorithms were recently utilised to identify and diagnose sick plants from digital photographs, automating plant disease diagnosis and aiding laypeople in recognising damaged plants. While several deep learning techniques are being utilised to identify ill plants and enhance detection rates, the primary constraint of model parameter size remains (Amin et al., 2022).
Early sickness identification can surpass current detection methods. Deep learning approaches based on computer vision, for example, might be used to detect diseases earlier. This article examines the disease classification and detection methodologies required to identify tomato leaf diseases in depth. This essay also assesses the benefits and drawbacks of the techniques presented. Finally, using a hybrid deep learning architecture, our study proposes an early disease detection technique for detecting tomato leaf diseases (David, 2021).