Keywords

Convolutional Neural Networks, Nutrient Deficiency Diagnosis, Computer Vision, Mechanical Design, Image Analysis, Machine Learning

Abstract

Strawberry plants exhibit various nutrient deficiency symptoms on their leaves. The discoloration and pattern can be used to detect and classify nitrogen, potassium, and phosphorous deficiencies. Given the effectiveness and efficiency of Convolutional Neural Networks (CNNs) in classification tasks, we chose several pretrained CNN architectures and used them to fit our custom image dataset. After compiling and labeling a dataset of 15,173 strawberry leaf images, it was divided into training and test sets, which were subsequently evaluated on several CNN models. The following model architectures were selected based on the performance and computational cost: ResNet50 V2, EfficientNetB0, MobileNet V3 Small, and MobileNet V3 Large. After training and evaluation, the best performing trained model achieved the target accuracy of 90% and was deployed onto a custom designed prototype for nutrient deficiency prediction in strawberry fields. This study details methodology, findings, and implications of a novel diagnostic tool aimed at assisting strawberry growers in optimizing crop health and yield.

Completion Date

2024

Semester

Spring

Committee Chair

Xu, Yunjun

Degree

Master of Science in Mechanical Engineering (M.S.M.E.)

College

College of Engineering and Computer Science

Department

Mechanical and Aerospace Engineering

Degree Program

Mechanical Engineering

Format

application/pdf

Language

English

Rights

In copyright

Release Date

11-15-2029

Length of Campus-only Access

5 years

Access Status

Masters Thesis (Campus-only Access)

Campus Location

Orlando (Main) Campus

Accessibility Status

Meets minimum standards for ETDs/HUTs

Download

Restricted to the UCF community until 11-15-2029; it will then be open access.

Share

COinS