Methods for estimating the weight of a fish.

Abstract

A method for estimating the weight of a fish by first using a camera to take a picture of the fish and sending the picture to at least one AI model, wherein the picture is classified in terms of what species was caught, then the picture is segmented by setting all pixels containing the fish to a first value and the remaining pixels to a second value, the segmented picture is sent to an algorithm to find the fish's centerline, all pixels on the calculated centerline are saved, the pixels on the calculated centerline are mapped to "real world" coordinates (x, y, z), then augmented reality (AR) points (x, y, z) are placed along the calculated fish's centerline, all points along the calculated centerline are added (integrated) to estimate the fish's body length, and empirical body length and weight values ​​of the species are used to estimate the fish's weight based on the estimated body length and species of the fish.

Description

Technical Field

[0001] The present invention relates to a method for estimating the weight of fish.

Background Art

[0002] There is no method that can analyze fish from a photo taken with a mobile phone or the like and register what kind of fish it is, its weight, body length, and where and when it was caught.

[0003] The invention described in the present application solves the above problems according to the claims.

Summary of the Invention

Means for Solving the Problems

[0004] The present invention is a method for estimating the weight of fish, comprising: - taking a photo of the fish using a camera; - sending the photo to at least one AI model, wherein the photo is classified as to what fish species was caught; - segmenting the photo by setting all pixels including the fish to a first value and the remaining pixels to a second value; - sending the segmented photo to an algorithm to find the center line of the fish and saving all pixels on the calculated center line; - mapping the pixels on the calculated center line to "real world" coordinates (x, y, z); - then arranging augmented reality (AR) points (x, y, z) along the calculated center line of the fish; - summing (integrating) all points along the calculated center line to estimate the body length of the fish; - using empirical body length and weight values of the species to estimate the weight of the fish based on the estimated body length and species of the fish.

[0005] Further embodiments are defined in the appended dependent claims. [Brief explanation of the drawing]

[0006] [Figure 1] This diagram shows the procedure for segmenting a fish photograph. [Modes for carrying out the invention]

[0007] One embodiment of the present invention includes the following steps: - Task flow on mobile phones.

[0008] 1. The user takes a picture of a fish with their mobile phone. The fish must be positioned so that it fills as many pixels as possible in the picture (the more pixels the fish occupies, the better the result).

[0009] At the same time, depth images from the AR are saved.

[0010] 2. Next, the photo is sent to two AI models. a. The photos are categorized to determine what species was caught. b. The photo is segmented. All pixels containing the fish are set to a value of 1, and the rest are set to a value of 0.

[0011] 3. The segmented images are then sent to a proprietary "skeleton" algorithm to find the fish's centerline. All pixels on the calculated centerline are saved.

[0012] 4. Now that we have the calculated centerline of the fish, we map these pixels to "real-world" coordinates (x, y, z). Next, we place Ar points (x, y, z) along the calculated centerline of the fish. Finally, we add (integrate) all the points along the curve to obtain an estimate of the fish's body length.

[0013] 5. The estimated body length and species are known. The inventors use empirical body length and weight values ​​(k coefficient) of the species to derive an estimate of the fish's weight.

[0014] AI fish classification and segmentation TensorFlow Lite is a lightweight and optimized version of the TensorFlow machine learning framework designed to run machine learning models on mobile and embedded devices, as well as in other resource-constrained environments. It is a software library that enables developers to deploy machine learning models on mobile devices, including smartphones, tablets, and even microcontrollers.

[0015] TensorFlow Lite is designed to deliver high performance and low latency, enabling it to run efficiently on devices with limited processing power and memory. This is achieved by using techniques such as model quantization, which reduces the precision of model weights and biases, allowing them to be stored and processed using fewer bits. TensorFlow Lite also supports hardware acceleration on a variety of platforms, including CPUs, GPUs, and dedicated machine learning hardware such as Google's Edge TPU. In addition to providing a runtime for running machine learning models on devices, TensorFlow Lite also includes tools for converting models trained with the full TensorFlow framework into a format usable with TensorFlow Lite. This makes it easy for developers to deploy TensorFlow-trained models on mobile and embedded devices.

[0016] Overall, TensorFlow Lite makes it easier for developers to bring machine learning capabilities to mobile and embedded devices, enabling them to build intelligent applications that run locally on the device without requiring network connectivity or cloud-based processing.

[0017] "Real World Location" of AR ARCore is a software development kit (SDK) developed by Google, which provides developers with tools to build augmented reality (AR) experiences on Android devices. ARCore uses the cameras and sensors on the device to understand the real world and interact with it, enabling users to place virtual objects and information on top of the actual environment. ARCore also supports features such as motion tracking, environmental recognition, and light detection to provide a more realistic and engaging AR experience.

[0018] ARKit is a similar SDK developed by Apple to build augmented reality (AR) experiences on iOS devices.

[0019] ARKit also uses the device's cameras and sensors to understand the real-world environment and interact with it, providing developers with tools to place virtual objects and information on top of it. ARKit also supports features such as motion tracking, environmental recognition, and light detection, as well as SceneKit and SpriteKit, to provide a more advanced AR experience. Both ARCore and ARKit help make augmented reality more accessible and easier to use for mobile device developers and users.

[0020] Skeleton The topological skeleton, also known as topological thinning, is a technique used in image processing and computer vision to simplify an object and represent it as a series of lines or curves.

[0021] The topological skeleton is composed of thin, robust shape representations that maintain its geometric shape and structure while eliminating unnecessary details.

[0022] The topological skeleton can be calculated using an algorithm that iteratively thins an image or shape by removing pixels or voxels that are not part of the skeleton. As a result, the object is represented as a series of points that represent the center of the original shape or line.

[0023] The topological skeleton can be used in various image processing and computer vision applications, such as the analysis of biological tissues, the recognition of objects and patterns, and the segmentation of images and patterns, as well as for image segmentation. It may also be useful for reducing the amount of data required to represent an object, which can be beneficial in applications where memory or processing power is limited.

[0024] Result Overall, this approach combines the power of AI and computer vision with the user-friendly interface of AR to provide a convenient and accurate way to measure the body length of fish in the field.

[0025] The following shows the segmentation and skeleton algorithms.

[0026] Skeleton #1 : Zhang's method (A fast parallel algorithm for thinning digital patterns, T.Y. Zhang and C.Y. Suen, Communications of the ACM, March 1984, Volume 27, Number 3).

[0027] Skeleton #2 :Lee's method (T.-C.Lee, RLKashyap and C.-N.Chu, Building skeleton models via 3-D medial surface / axis thinning algorithms. Computer Vision, Graphics, and Image Processing, 56(6):462-478, 1994).

[0028] Custom Skeleton #1: In-house Development (Fiskher)

Claims

1. A method for estimating the weight of a fish, - The step of taking a photograph of the fish using a camera, - A step of transmitting the aforementioned photograph to at least one AI model, wherein the photograph is classified in terms of what species was captured, - The step of segmenting the photograph by setting all pixels containing the fish to a first value and the remaining pixels to a second value, - The steps include sending the segmented photograph to an algorithm to find the centerline of the fish and saving all pixels on the calculated centerline, - A step of mapping the pixels on the calculated center line to "real world" coordinates (x, y, z), - Next, the augmented reality (AR) points (x, y, z) are positioned along the calculated centerline of the fish. - A step of estimating the body length of the fish by summing (integrating) all points along the calculated center line, A method comprising the step of using empirical body length and weight values ​​of the species to estimate the weight of the fish based on the estimated body length and species of the fish.

2. The method according to claim 1, wherein the camera is connected to a mobile device such as a mobile phone or tablet.

3. The method according to claim 1, wherein the empirical values ​​of body length and weight are the k coefficient.

Images