DESIGN AND EVALUATION OF CONVOLUTIONAL NETWORKS FOR VIDEO ANALYSIS OF BEE Traffic is a well-researched Physical Sciences and Mathematics Thesis/Dissertation topic, it is to be used as a guide or framework for your Academic Research.
This thesis presents a novel approach to estimate bee traffic levels in a Langstroth beehive by employing Deep Learning and computer vision techniques to recognize bees. Estimating bee traffic levels will further help in analyzing bee colony behavior and health. Various Convolutional Neural Networks (ConvNet) models were designed, trained, and validated to find the best model for bee recognition.
The best model is then integrated with a motion detection algorithm running on a Raspberry Pi 3 computer as a part of Beepi, a multi-sensor electronic beehive monitoring system, that was designed in 2014 and has been iteratively developed ever since.
We compared the bee detection accuracy of ConvNets with traditional machine learning models and found ConvNets to be superior. Best performing models were then used to estimate bee traffic levels, aligned with real Beehive inspections. This was further extended to find a correlation between the Beehive data points extracted from inspections and be traffic levels.
The Western honey bee (Apis mellifera L.) is the world’s premier pollinator species. Due to soaring demands for fruits and especially nuts, their importance as a pollinator has increased as well. That represents almost 100 crop species, making up one-third of the average diet. In the United States, honey bees pollinate an estimated $15 billion of crops each year .
Apiculture has been on the decline  in both the USA and Europe over recent decades. It is, therefore, crucial to make beekeeping a more attractive hobby and a less laborious profession, in order to encourage local apiculture and pollination. Sudden losses of honey are colonies have occurred, and have received considerable public attention.
Colony Collapse Disorder (CCD) in the USA has attracted great attention, and scientists here and in Europe are working hard to provide explanations for these extensive colony losses. Disruption of the honey bee supply causes prices for domestically grown nuts, fruits, and vegetables to go up.
These factors have led bee health to be a cause of urgent concern for scientists and farmers around the world for at least a decade and a lot of work has been done in past for data collection and analysis for its research. One of the important variables in order to monitor food availability, food demand, colony age structure, the impact of pesticides, etc.
in bee-hives is forager traffic . Forager activity is an important variable to monitor when evaluating the impact of pesticides on honey bee colony health . Monitoring forger traffic will lead to improved remote monitoring of general hive status and improved real-time detection of the impact of pests, diseases, pesticide exposure, and other hive management problems.
Since forager traffic can be affected by food availability, food demand, and colony age structure. Thus, sudden changes in that traffic may indicate acute changes on the colony level.
Forager activity is described in terms of the number of bees entering and/or exiting the hive over a given time period, data can be collected, if need be, without the use of equipment more sophisticated than an observer and a stopwatch. The use of human observation, while likely to be accurate, clearly limits due to fatigue and the amount of time that the hive can be observed.
The importance of estimating the health of bee-hives demands research, data collection, and beehive setups over the span of several years. EBM systems can help in automating. the collective effort of a large amount of information to assess hive health without invasive and disruptive colony inspections.
With more upcoming and previous research work on such systems, EBMs have gained more popularity over the years. Since the usage of object detection methodologies plays a prime role in these systems, a good amount of high-quality work has been done in the field of object detection using machine learning.
The importance of estimating the health of bee colonies demands research and advancement over the span of several years in this area. EBM systems help in automating the task of collecting a large amount of useful information on behavior without invasive and disruptive colony inspections. As EBMs gained popularity, it became important to focus on the design of EBMs as well as incorporating modern software solutions to aid and better utilize them. Related work has been described below.